negative shocks and mass persecutions: evidence from the...

Negative Shocks and Mass Persecutions:Evidence from the Black Death

Remi Jedwab and Noel D. Johnson and Mark Koyama⇤

April 16, 2018

Abstract

We study the Black Death pogroms to shed light on the factors determining when a

minority group will face persecution. In theory, negative shocks increase the likelihood that

minorities are persecuted. But, as shocks become more severe, the persecution probability

decreases if there are economic complementarities between majority and minority groups.

The effects of shocks on persecutions are thus ambiguous. We compile city-level data

on Black Death mortality and Jewish persecutions. At an aggregate level, scapegoating

increases the probability of a persecution. However, cities which experienced higher plague

mortality rates were less likely to persecute. Furthermore, for a given mortality shock,

persecutions were less likely in cities where Jews played an important economic role and

more likely in cities where people were more inclined to believe conspiracy theories that

blamed the Jews for the plague. Our results have contemporary relevance given interest in

the impact of economic, environmental and epidemiological shocks on conflict.

JEL Codes: D74; J15; D84; N33; N43; O1; R1

Keywords: Economics of Mass Killings; Inter-Group Conflict; Minorities; Persecutions;

Scapegoating; Biases; Conspiracy Theories; Complementarities; Pandemics; Cities

⇤Corresponding author: Remi Jedwab: Associate Professor of Economics, Department of Economics, GeorgeWashington University, [email protected]. Mark Koyama: Associate Professor of Economics, Department ofEconomics, George Mason University, [email protected]. Noel Johnson: Associate Professor of Economics,Department of Economics, George Mason University, [email protected]. We are grateful to Sascha Becker, MichaelBordo, Quoc-Anh Do, Jose-Antonio Espin-Sanchez, James Fenske, Raphael Franck, Francisco Gallego, SaumitraJha, Naomi Lamoreaux, Moses Shayo, Oliver Vanden-Eynde, Joachim Voth, Ekaterina Zhuravskaya, and seminaraudiences at American, ASREC, Ben Gurion, Chapman University, CIREQ Conference (Universite de Montreal), theEHS Meetings (Royal Holloway), Florida State, George Mason, George Washington, Hebrew, Hitsubashi, PontificiaUniversidad Catolica de Chile, Rosario, Rutgers, Sciences Po, the SEA Meetings (Washington D.C.), Sussex, theWashington Area Economic History Workshop (George Mason), the Washington Area Development EconomicsSymposium (George Mason) and Yale. We gratefully acknowledge the support of the Institute for InternationalEconomic Policy at George Washington University and the Mercatus Center at George Mason University.

http://home.gwu.edu/~jedwab/

http://noeldjohnson.net/noeldjohnson.net/Home.html

http://mason.gmu.edu/~mkoyama2/About.html

NEGATIVE SHOCKS AND MASS PERSECUTIONS: EVIDENCE FROM THE BLACK DEATH 1

What factors make minorities vulnerable to persecution? We answer this question by studyingJewish communities during the Black Death, the greatest demographic shock in European history:about 40% of the population died between 1347 and 1352 (Benedictow, 2005; Christakos et al.,2005). We use data on city-level Black Death mortality rates and Jewish persecution to show thatthe higher mortality was in a city, the less likely was a persecution to occur. Furthermore, wefind this effect was accentuated in places where Jews played an important economic role andattenuated in cities where people were inclined to believe Jews had caused the plague.

Understanding the causes of the Black Death persecutions is important in its own right. Thesehave been described as “the most monumental of medieval Jewish persecutions” (Cohn, 2007)and “precursor(s) of the Holocaust” (Goldhagen, 2013). Studying this episode is also importantbecause it sheds light on other episodes of intergroup conflict. Minorities remain targets ofviolence in many parts of the world (Chua, 2004; Field, Levinson, Pande and Visaria, 2008; Jha,2013; Yanagizawa-Drott, 2014). But why are minorities protected in some places but not in others?1

We consider a simple framework to analyze how a negative shock affects the incentive of aningroup, to persecute a minority or outgroup. In our framework, the outgroup community providesimportant economic services but can be held responsible for the shock of the plague. We show thatas the severity of the shock increases, there are two potentially conflicting effects. If the economicimportance of the outgroup increases as the shock worsens, then there is a greater incentive toprotect the outgroup following a more severe shock. This is the complementarities effect. If theoutgroup is blamed for the shock, however, then the incentive to persecute the outgroup canincrease with the severity of the shock. This is the scapegoating effect.

We then compile city-level data on the virulence and spread of the Black Death along withinformation on the locations and characteristics of Jewish communities and persecutions acrossWestern Europe. These data cover both the Black Death period (1347-1352) and surroundingcenturies (1100-1850). Among the 1,869 cities in Western Europe, 363 had a Jewish communityat the onset of the Black Death. We have estimates of Black Death mortality rates for 263 localities.Our main sample is the intersection of these two datasets, which consists of the 124 cities thathad a Jewish community in 1347 and for which we know their mortality rates. This sampleis representative for Western Europe: the 124 cities in our dataset represent 66.7% of the totalpopulation of all cities with a Jewish community and 43% of the total urban population in 1300.

About half of cities with a Jewish community suffered a persecution during the Black Death.However, cities with higher mortality rates were less likely to persecute their community.

We believe these results are causal. (i) We provide evidence that the virulence of the plaguewas plausibly unrelated to factors related to persecutions. (ii) The parallel trends assumption

1Many minority groups across the world still suffer violent persecutions today, with some recent examples of ethnicand religious cleaning in the Central African Republic, Myanmar, South Sudan, Sudan, Sri Lanka, and Syria.

2 REMI JEDWAB, NOEL D. JOHNSON AND MARK KOYAMA

is verified as, prior to 1347, there was no difference in persecutions between the cities mostaffected and those comparatively unaffected by the Black Death. (iii) Results are robust to theinclusion of geographic and institutional controls, as well as controls for community size and pastpersecutions. (iv) Our results hold when we implement an instrumental variables (IV) strategypremised on the fact that the Black Death entered Europe through the Sicilian port of Messinaand that it was more virulent early on. The first IV we construct is market access to Messinaconditional on market access to all cities, as it was the specific connectedness to Messina and notthe connectedness to other important trading cities that mattered for plague virulence. The secondIV is based on the number of months between the year-month the city was infected and October1347 (the date of first contact). This last IV relies on the randomness of the timing of infection.

We shed light on the pathways that explain both why the overall level of persecution was highand why cities with higher mortality rates persecuted less. We find that, conditional on thesize of the mortality shock, Jews were more likely to be persecuted in towns where people wereinclined to believe antisemitic canards (scapegoating effect). Starting with the First Crusade (1096),persecutions were increasingly perpetrated against Jews. From the 12th century onwards Jewswere accused of ritually murdering Christian children. During the Black Death Jews were torturedinto confessing to have caused the plague by poisoning wells. We find that the protective effectof high mortality was attenuated for towns closer to where such accusations were made. Theprotective effect was also weaker in cities first infected during Christmastide and Easter — whenChristians historically blamed Jews for the death of Jesus — and stronger for Advent and Lent —when Christians were doing penance.

Conversely, Jews were less likely to be persecuted at higher mortality rates in cities where theycould offer specialized economic services (complementarities effect). Conditional on the size of theshock, we find a lower probability of persecution in larger and better connected cities and in citieswhere Jews were offering moneylending services or services to the trading sector. In addition,cities with a community grew relatively faster than cities without, and cities that persecuted theircommunity during the Black Death grew relatively slower in the following centuries.

We contribute to the literature on the economic determinants of minority persecution. First, thereare only a few studies on the economics of mass killings (Easterly et al., 2006; Montalvo andReynal-Querol, 2008; Caselli et al., 2015; Esteban et al., 2015; Rogall and Yanagizawa-Drott, 2016).Arbath, Ashraf and Galor (2015) show that genetic diversity, as determined predominantly tensof thousands of years ago, can explain the frequency, incidence, and onset of civil conflicts andinterethnic violence. Esteban et al. (2015) explain that the decision by a majority to eliminate itsminority is the outcome of strategic and rational economic calculation: The majority group startsa genocide if the net present value of expected pay-offs is positive, which depends on the size ofthe winnable surplus and the productivity losses from conflict. The Black Death pogroms alsotook place in a Malthusian setting (Galor, 2005, 2011). This makes them particularly relevant forstudying outbreaks of ethnic or religious violence in African countries whose economies may have


remained broadly Malthusian. Indeed, following the Rwandan genocide, Rogall and Yanagizawa-Drott (2016) find that living standards are higher in villages that experienced greater violence.There is also a literature on the effects of negative shocks (e.g., climate shocks) on the persecutionof minorities (Oster, 2004; Miguel, 2005; Anderson et al., 2016). Studying Czarist Russia, Grosfeld,Sakalli and Zhuravskaya (2017) find that pogroms were more likely during a hot growing seasonif Jews were creditors or grain traders. Economic complementarities between groups can shapethe vulnerability of a group to conflict (Jha, 2007, 2013, 2018). Jha (2013) finds that toleration ofMuslims in, majority Hindu, Indian cities depends on the complementarity/substitutability of theservices they provide. Jha (2018) models the conditions under which trade can support peacefulcoexistence between groups and those under which groups will be subject to violence. Becker andPascali (2016) study post-Reformation Germany and show that persecutions were more likely inProtestant than in Catholic areas as it was in the former that there were individuals who couldsubstitute for the role of moneylenders that Jews had previously played.2

Second, there are few studies on the mechanisms through which scapegoating spreads after ashock. We document how biases and conspiracy theories conditioned the effect of mortality onpersecution probability. Our paper thus contributes to the literature on the role of disinformation(Gentzkow and Shapiro, 2006; DellaVigna and Kaplan, 2007; Enikolopov et al., 2011; Vigna etal., 2014; Adena et al., 2015; Allcott and Gentzkow, 2017), and more generally propaganda(Yanagizawa-Drott, 2014; Voigtlander and Voth, 2014, 2015; Bonnier et al., 2015; Rogall, 2015).Given today’s information technologies, “fake news” can spread across space extremely fast,which makes it difficult to measure its local effect. By focusing on the gradual spread of fakenews in the medieval era, we can identify its local effects. Remarkably, accusations of Jewishdeicide and blood and well-poisoning libels are still very relevant today.3

Third, we add to the literature on the economics of epidemics. While this literature has focusedon understanding why pandemics spread (e.g. Oster, 2005; Boerner and Severgnini, 2014; Dufloet al., 2015), there are fewer studies of their economic consequences (Young, 2005; Almond, 2006;Voigtlander and Voth, 2013b). To our knowledge, there are no studies on the effects of pandemicson minority persecution, such as in the 1980s when HIV spread in the U.S., resulting in a wave of

2Our paper is also related to the literature on ethnic diversity and conflict (Montalvo and Reynal-Querol, 2005;Esteban and Ray, 2008, 2011; Esteban et al., 2012; Caselli and Coleman, 2013; Rohner et al., 2013a,b; Arbatli et al.,2015; Ray and Esteban, 2016; Mayoral and Ray, 2016; Michalopoulos and Papaioannou, 2016). This literature showsthat conflict takes place along ethnic lines, because: (i) ethnicity is not easily disguisable (Caselli and Coleman,2013); (ii) conflict requires both economic resources and conflict labor, which can both be found within a same ethnicgroup but not within a same class (Esteban and Ray, 2008); and (iii) ethnicity proxies for income if ethnic groups areprofessionally specialized (Esteban and Ray, 2011). Our context fits these characteristics, as Jews in medieval Europewere economically specialized and often wore distinctive clothing.

3For example, the Pizzagate conspiracy theory that emerged during the 2016 U.S. presidential election alleged that“Washington progressives [. . . ] dine on satanic food watered with the blood of babies [. . . ] The underpinnings ofthe baby’s blood allegations will be horrifyingly familiar to Jews who grew up hearing about it from family memberswho escaped European pogroms” as it references the “centuries-old false allegation that Jews murder [. . . ] Christianchildren to use their blood for ritual purposes” (Newsweek, 2016). The New York Times (2016) also reported: “Echoinganti-Semitic claims that led to the mass killings of European Jews in medieval times, President Mahmoud Abbas of thePalestinian Authority accused rabbis in Israel of calling on their government to poison the water used by Palestinians.”


attacks on the homosexual community (New York Times, 1986). Conspiracy theories also emergewhen epidemics break out, for example after the West African Ebola epidemic of 2013-2016 (NewYork Times, 2014). Understanding the effects of pandemics is all the more important given that theoccurrence of such events will increase with climate change, which “is melting permafrost soilsthat have been frozen for thousands of years [. . . ] releasing ancient viruses” (BBC, 2017).4

Lastly, we contribute to the literature on antisemitism. In an important contribution, Voigtlanderand Voth (2012) use data on the Black Death pogroms (but not on Black Death mortality rates)to explore the persistence of antisemitic cultural traits from the 14th century through to the20th century in Germany. The bulk of research has focused on the economic consequences ofantisemitism(Waldinger, 2010, 2012, 2016; Acemoglu et al., 2011; Grosfeld et al., 2013; Pascali,2016; D’Acunto et al., 2017). We also explore the local persistence of antisemitic cultural traits,from the 11th century through to the 14th century across all of Western Europe, and find that theBlack Death persecutions were associated with slower city growth in the long run.5

Our study has several novel features as Europe during the Black Death provides an uniquelywell suited setting to explore the causes of minority persecutions. We have many cities, in morethan one country, with two well-identified groups. We exploit a shock that was massive, highlyvariable, and plausibly random. The IV strategies are based on market access and epidemiologicalfactors, two relatively novel sources. Our context of decentralized polities with weak statecapacity is similar to that observed in poor countries today, where minorities often face violence.However, such persecutions are rarely studied, due to lack of data. As our data cover all of Europe,we exploit the rich variation in initial conditions across cities and study how the shock-persecutionrelationship varies with economic and other social institutions, allowing us to study the effects ofboth scapegoating and complementarities. In contrast, many studies are constrained to focus ononly one set of institutions and one type of effect at a time due to more limited data.

1. Conceptual Framework

We now describe how a shock can affect the probability that an outgroup is persecuted. Percapita income y depends on population L. In line with the historical discussion, it also dependspositively on the value of the economic services provided by the minority x: y = f(L, x).

The utility of in-group members has three components: per capita income y, the death of in-groupmembers �, and preferences over diversity z (i.e. preferences over the out-group):

u = g(y, �, z)

4The Black Death killed about 40% of Europe’s population. By comparison, the 1918 flu pandemic killed 75 millionpeople, or 4% of the world’s population. No diseases have been as deadly in the 21st century, although HIV infectionrates exceed 20% in Southern Africa. Nevertheless, many influential actors in the development community think thatinfectious disease pandemics are the biggest threat to humanity, one example being Bill Gates (see his 2015 article “TheNext Epidemic — Lessons from Ebola” published in the The New England Journal of Medicine (372:1381-1384)).

5Relatedly, there is a literature on the origins of Nazism and the persistence of antisemitic attitudes in Germany(Voigtlander and Voth, 2012, 2013a, 2014, 2015; Adena et al., 2015; Satyanath et al., 2017).


For simplicity, we work with a linear, separable, version of this function (thereby ignoring cross-effects). The effect of the Black Death mortality M on the utility of a representative member of thein-group (who survives the plague) depends on how these three components are affected:

du

dM

=

dy

dM

+

d�

dM

+

dz

dM

(1)

The first component, the effect of the Black Death shock on per capita income could be positivedue to the Malthusian effect on per capita income ( dL

dM@y@L ). However, as we will describe below, the

Black Death disrupted production causing incomes to initially decrease. The second component( d�dM ) is negative: the death of family, friends and others reduces utility. The third effect is context

specific. If the minority is held responsible for the shock, then it is negative ( dzdM ), and the in-group

resents the presence of the out-group more, the larger the shock.

We can solve for the decision to persecute the outgroup (see Web Appendix Section 1 for details).6

We show that a representative individual will persecute the outgroup if the scapegoating effect islarger than the complementarities effect:

��dL

dM

⇣@y(L, x)

@L

� @y(L)

@L

⌘

| {z }Complementarities Effect

�� <��

dz

dM|{z}Scapegoating Effect

�� . (2)

If relative size of the complementarity and scapegoating effects vary from city to city then a givenmortality shock could have differential effects. Moreover for a given set of city characteristics,the size of shock might give rise to differential effects. If the economic role of the minoritybecomes more important in the presence of a large shock, this can generate an inverted U-shapedrelationship between Black Death mortality and the probability of a persecution.

We proxy for the scapegoating and complementarities effects in our analysis using characteristicsof the towns. We will proxy for the strength of the scapegoating effect using data such as howclose a town was to the origin of rumors of well poisoning by Jews, whether the town had ahistory of persecuting Jews, and the timing of the arrival of the Plague (e.g. Easter). Similarly, wewill proxy for the strength of the complementarities effect using measures such as whether Jewswere lending money in the town and whether Jewish traders had access to more markets.

Specifically, we will be able to show that in the presence of scapegoating characteristics a town wasmore likely to persecute the Jewish community, holding constant the mortality rate. By contrast,when there was evidence of economic complementarities between Jews and non-Jews, then thereis a strong, inverse relationship, between mortality rates and persecution probability.

6We find the level of utility at which an individual is indifferent between persecuting and not persecuting theoutgroup. To see how this is affected by plague mortality we employ the method of comparative statics and takethe total derivative of this condition with respect to mortality M .


2. Data

This section presents our data (see the attached Web Appendix for more details).

Black Death Mortality. Data on cumulative Black Death mortality for the period 1347-1352 comefrom Christakos et al. (2005, 117-122) who compile data on mortality rates based on informationfrom a wide array of historical sources including ecclesiastical records, parish records, testaments,tax records, court rolls, chroniclers’ reports, donations to the church, financial transactions,passing away of famous people, letters, edicts, guild records, hospital records, cemeteries andtombstones. Christakos et al. (2005) carefully examine each data point and arbitrate betweenconflicting estimates based on the best available information. We have checked these data byconsulting numerous other sources including Ziegler (1969), Russell (1972), Gottfried (1983), andBenedictow (2005). These data yield estimates of mortality for 263 localities in 13 countries. Figure1 depicts these 263 localities and their mortality rates.

For 166 of these 263 localities we have a percentage estimate of the mortality rate. For example,Venice had a mortality rate of 60%. In other cases the sources report more qualitative estimates:(i) For 49 towns Christakos et al. (2005) provide a literary description of mortality. We rank thesedescriptions based on the implied magnitude of the shock and assign each one of them a numericmortality rate.7 (ii) For 19 towns we know the mortality rate of the clergy. Christakos et al.(2005) provide evidence that clergy mortality was on average 8% higher than general mortality,so we divide the clergy mortality rates by 1.08 to obtain their mortality rate. (iii) For 29 townswe know the desertion rate, which includes both people who died and people who never cameback. Following Christakos et al. (2005, 154-155), who show that the desertion rate is on average1.2 times higher than the mortality rate, we divide desertion rates by 1.2 to obtain their mortalityrate. Since the data generated for the towns of (i), (ii) and (iii) require us to make assumptions, wewill show that our results are robust to using only numerical estimates.

Jewish Presence, Jewish Persecution, and Main Sample. We use the Encyclopedia Judaicawhich provides comprehensive coverage on Jewish life to determine which towns had Jewishcommunities (Berenbaum and Skolnik, 2007a). The Encyclopedia Judaica provides details on Jewishpresence, economic activities, and persecutions. The length of entry varies: some are rich withhistorical information, others are shorter. An example entry is from Toulon.

In the second half of the 13th century the Jews made up an appreciable proportionof the population of Toulon: at a general municipal assembly held in 1285, 11 of the155 participants were Jews. [. . . ] The community came to a brutal end on the night ofApril 12/13, 1348 (Palm Sunday), when the Jewish street, “Carriera de la Juteria,” wasattacked, the houses pillaged, and 40 Jews slain; this attack was probably related to theBlack Death persecutions. [. . . ] After this date, in addition to a few converted Jews,

75% for “spared” or “escaped”, 10% for “partially spared” or “minimal”, 20% for “low”, 25% for “moderate”, 50%for “high”, 66% for “highly depopulated”, and 80% if the town is “close to being depopulated” or “decimated”.


there were in Toulon only individual Jews who stayed for short periods.

We supplement these data where possible with other sources including the Jewish Encyclopedia(Adler and Singer, eds, 1906). Among the 1,869 Western European towns that reached 1,000inhabitants at one point between 800 and 1850 in the Bairoch (1988) database and/or belong tothe Christakos et al. (2005) database of mortality rates, we have identified 363 towns in whichJews were present circa the onset of the Black Death in 1347. Of these 363 Jewish communities, wecan match 124 locations to our database of mortality rates.

Our main sample thus consists of 124 towns with a Jewish community at the onset of the BlackDeath (circa 1347) and for which we know the Black Death mortality rate (%, 1347-1352). Figure 2depicts these 124 towns. In addition, we also know for the full sample of 1,869 towns which townhad a Jewish community in each year from 1100 to 1850. Our sample is representative: The 124towns account for 66.7% of the total population of the 363 towns with a Jewish community at theonset of the Black Death, and 43% of the total population of the 1,869 towns in 1300.

Note that data do not exist on the size of Jewish communities in the 14th century. For 30 out of the124 towns, however, we do have estimates of the share of Jews in the population of the town.

Our dependent variable is whether a community was persecuted. We focus on the Black Deathperiod (1347-1352), but we have these data for each town with community from 1100 to 1850. Ourdefinition of a persecution encompasses both a pogrom (an act of violence) or an expulsion. Forour main sample, and during the Black Death period only, we use additional information on thedate of the persecution, the number of victims, whether the community was annihilated, whetherJews were burned, and whether a mob was involved in the persecution.

Black Death Spread. We use the data from Christakos et al. (2005) to obtain for 95 of the 124towns the year and month of first infection. For the other 29 towns, we rely on information forneighboring towns, maps of the epidemic spread, and extra sources to impute the year-month offirst infection.8 Information is sparser for the year-month of last infection.

Controls. Geographic controls include mean growing season temperature in 1500-1600 (nocomparable data before), elevation, soil suitability for cereal production and pastoral farming,dummies for whether the town is within 10 km from a coast or river, and longitude and latitude.9

To control for factors related to trade, we employ data on populations in 1300 from Jebwabet al. (2016), who combine data from both Bairoch (1988) and Chandler (1987). The last twosources represent attempts to collect information on populations for all towns with at least 1,000inhabitants. For the towns for which population is not available, we believe that it must be

8For example, for Landshut in Germany we learn from Benedictow (2005, 190) that the epidemic went fromMuhldorf to the neighboring town of Landshut (50 km). From Christakos et al. (2005), we know that Muhldorf andRegensburg were first infected in June and July 1349, respectively. Since Landshut is about one-half of the way betweenMuhldorf and Regensburg, it must have been infected in June or July 1349, but most probably in June 1349.

9We use 10 km for the coast and river dummies and some controls described below, due to measurement error.


less than 1,000 and thus arbitrarily assume that their population was 500. We also control forthe presence of major and regular Roman roads (and their intersections) using the GIS datafrom McCormick et al. (2013), medieval trade routes (and their intersections) after digitizing amap from Shepherd (1923), and two dummies capturing the presence of medieval market fairsand membership in the Hanseatic league based on information from Dollinger (1970). We alsocalculate market access for every town to the 1,869 towns of the full sample in 1300. Market accessfor town i is defined as MAi = ⌃j

Lj

⌧�ij, with Lj being the population of town j 6= i, ⌧ij the travel

time between town i and town j, and � = 3.8, as in Donaldson (2018). We compute the least costtravel paths via four transportation modes—sea, river, road and walking—using the data fromBoerner and Severgnini (2014) who estimate the speed at which the Black Death traveled.

Our human capital controls include a dummy for whether a town possessed a university (Boskeret al., 2013) and for whether a town was within 10 km from a Roman aqueduct (Talbert, ed, 2000).To control for institutions, we distinguish between towns that were located in monarchies andself-governing cities around 1300 (Bosker et al., 2013; Stasavage, 2014). We control for whether thetown was a state capital around 1300. We measure parliamentary activity during the 14th centuryusing data from van Zanden et al. (2012) and control for whether a city was within 100 km of abattle that took place between 1300 and 1350.

3. Historical Setting

The Black Death. The Black Death arrived in Europe in October 1347. Over the next few years itspread across the continent killing between 30% and 50% of the population.10 Recent discoveriesin plague pits have corroborated the hypothesis that the Black Death was Bubonic plague. Thebacterium Yersinia Pestis was transmitted by the fleas of the black rat. Infected fleas suffer froma blocked esophagus. These “blocked” fleas are unable to sate themselves and continue to biterats or humans, regurgitating the bacterium into the bite wound thereby infecting rats or humans.Within less than a week, the bacteria is transmitted from the bite to the lymph nodes producingbuboes. Once infected, death occurred within ten days with 70% probability.11

While the vector for bubonic plague is infected fleas, fleas cannot spread the disease far in theabsence of black rats. The spread of the plague was rapid and its trajectory was largely determinedby chance. One important means of transmission depended on which ship became inhabitedwith infected fleas. It was largely coincidence that the plague spread first from Kaffa in the BlackSea to Messina in Sicily in October 1347 rather than elsewhere as the ships carrying the plaguewere originally bound to Genoa (but the ships could have been bound to other Mediterraneanports). Figure 1 shows the locations of Messina, Kaffa, and Genoa. Similarly, it was coincidentalthat the plague spread first from Messina to Marseilles, rather than to, say, Barcelona, Lisbon, or

10Conventionally the overall death rate was estimated at 1/3. Recent studies suggest that the death rate was higherthan this (see Benedictow, 2005, 2010; Aberth, 2009). For the 124 towns, the population-weighted average is 38.0%.

11See Benedictow (2005, 2010). The importance of blocked fleas as the main vector of transmission is currently underdebate. Other vectors such as lice may also have been at work. The literature agrees, however, that person-to-persontransmission was probably rare and cannot account for the diffusion of the plague (Campbell, 2016, 235).


Antwerp. From the various coasts where infected ships docked the plague then spread inlandalong rivers, roads, and paths, as boats and carts were inadvertently carrying infected fleas. Thelocal spread of the plague thus also depended on the local populations of black rats. Since blackrats are territorial, their numbers were not correlated with population density (Benedictow, 2005).For example, similar death rates are recorded in urban and in rural areas (Herlihy, 1965).

For mostly epidemiological reasons, virulence was greater in towns affected earlier on (Christakoset al., 2005, 212-213). Initially, epidemics spread exponentially. Eventually, epidemics run outof victims, which forces the disease to mutate, favoring benign pathogens.12 People may alsodevelop immunities and pathogen mutation may increase individual memory immune responsesdue to “contacted individuals becoming infected only if they are exposed to strains that aresignificantly different from other strains in their memory repertoire” (Girvan et al., 2002). In otherwords, pathogen mutation and natural immunization eventually cause an epidemic to end.

All towns were affected by the Black Death. Once an outbreak began in a town, mortality increasedrapidly before peaking two to three months after the date of first infection and then declining(Christakos et al., 2005, 212-213). Consistent with this, when we use available data on the year andmonth of first and last infection for 39 out of the 124 towns, the mean duration of the plague was5 months. Figure 3(a) shows the distribution of the duration of the epidemic for these.

Based on the historical and epidemiological facts described above, it is apparent that thespread, and thus the local virulence, of the plague had a significant random component. Whenstudying variation in mortality rates across space, historical accounts have been unable torationalize the patterns in the data (Ziegler, 1969; Gottfried, 1983; Theilmann and Cate, 2007;Cohn and Alfani, 2007). To illustrate, Venice had high mortality (60%) while Milan escapedcomparatively unscathed (15% mortality). Highly urbanized Sicily suffered heavily from theplague. Equally urbanized Flanders, however, had relatively low death rates. Southern Europeand the Mediterranean were hit especially hard, but so were the British Isles and Scandinavia.13

Consistent with this, Figure 4(a) illustrates the lack of a relationship between mortality ratesand city population in 1300. Likewise, Figure 4(b) shows that there is no relationship betweenmortality rates and city market access in 1300. Similarly, some scholars have argued that deathrates from the plague were lower in mountainous regions, but mortality rates in mountainousSavoy were high whereas “despite Switzerland having the most rugged terrain in Europe, theBlack Death reached almost every inhabited region of the country” (Christakos et al., 2005, 150).14

12According to Berngruber et al. (2013): “[. . . ] selection for pathogen virulence and horizontal transmission is highestat the onset of an epidemic but decreases thereafter, as the epidemic depletes the pool of susceptible hosts [. . . ] In theearly stage of an epidemic susceptible hosts are abundant and virulent pathogens that invest more into horizontaltransmission should win the competition. Later on, the spread of the infection reduces the pool of susceptible hostsand may reverse the selection on virulence. This may favor benign pathogens [. . . ].”

13Variation in sanitation does not explain this pattern. Gottfried (1983, 69) notes “it would be a mistake to attributetoo much to sanitation” given the “failure of Venice’s excellent sanitation to stem the deadly effect of the plague”.

14We also find no relationship with the density of population within walled cities (which allows for accurate densitycalculations). These data are available for 56 towns (Web Appendix Figure A.4).


The Black Death affected all segments of the population, rulers and commoners, rich and poor,adults and children, men and women. Neither the medical profession nor authorities were ableto respond effectively. Medical knowledge was rudimentary: Boccaccio (2005, 1371) wrote that“all the advice of physicians and all the power of medicine were profitless and unavailing”.Individuals, regardless of wealth, were unable to protect themselves. Institutional measures ofprevention were nonexistent: the practice of quarantine was not employed until 20 years later.15

To summarize: (i) The disease was more virulent initially, so towns that were randomly infectedearlier should have higher mortality rates; and (ii) The disease spread from Messina initially.Therefore, we expect towns that were better connected to Messina—but not necessarily betterconnected overall—to have higher mortality rates.

The Economic Role Played by Jewish Communities. In many parts of Europe, Jewishcommunities had settled in Roman times; elsewhere their settlement dated back to the Carolingianperiod (768-888) and thus predates our study by more than 5 centuries. Among the full sample of1,869 towns, 363 of them had a Jewish community in 1347, i.e. almost 20% of towns. Jews tendedto live in larger cities. The 124 towns of our main sample account for 66.7% of the total populationof the 363 towns with a Jewish community in 1347. The most important economic role playedby Jews at the time of the Black Death was as moneylenders which was due to their high levelsof human capital (Botticini and Eckstein, 2012) and the restrictions on lending money at interestissued by the Church (Koyama, 2010). Jews were also doctors and merchants. The value of Jewishcommunities was clearly perceived at the time (Chazan, 2010).16

The Black Death Persecutions. The Black Death led to economic collapse as fields lay fallow andtrade networks were disrupted. Prices rose due to this negative supply shock and this resultedin an immediate fall in incomes (Munro, 2004).17 The outbreak was accompanied by outbreaksof sexual and religious excess and by persecutions (Ziegler, 1969; Gottfried, 1983; Aberth, 2009).Figure 5(a) plots the persecutions in our dataset between 1100 and 1600. Figure 5(b) focuses on theBlack Death era (1347-1352), for our main sample. 58 out of the 124 towns – 47% of them – saw apersecution (53 pogroms and 13 expulsions). The majority of persecutions took place in 1348-1349.

Note that Jews were the only ethnic and religious minority in most towns. Protestantism did notbegin before 1517. Thus, with the exception of Narbonne where beggars were accused of havingpoisoned the wells, only Jews were persecuted during the Black Death period.

4. The Effect of Black Death Mortality on Jewish Persecutions15The term quarantine was first used in Ragusa, part of the Venetian empire in 1377 Gensini et al. (2004, 257).16Historians observe that “rulers concerned with attracting Jews offered promises of security and economic

opportunity to Jewish settlers” (Chazan, 2010, 103). This also appears in many entries of the Encyclopedia Judaica.17It was only years later that the labor shortage effect emphasized by Pamuk (2007) and Voigtlander and Voth (2013b)

was realized in rising wages. In addition, property rights were feudal and, as such, most property was owned by elites.While the Black Death did improve the bargaining position of peasants (e.g. Haddock and Kiesling, 2002; Acemogluand Robinson, 2012), it took decades for contractual obligations to be renegotiated.


4.1. Specification and Main Results.

We estimate a series of regressions based on:

Pi,1347�52 = ↵+ �Mi,1347�52 + "i , (3)

where Pi,1347�52 is a dummy equal to one if there is a persecution in town i between 1347-1352, andMi,1347�52 is the cumulative Black Death mortality rate (%) in town i over the period 1347-1352.

As discussed in Section 3., the plague lasted on average 5 months in each town. Thus, mortalityin 1347-1352 measures cumulative mortality over a period of a few months only (and not 5 years).Cumulative mortality was strongly correlated with monthly mortality, which increased quicklyafter the first infection before decreasing to 0. With a plague duration of 5 months, monthlymortality most likely peaked after 2 months. Most persecutions also took place during these 5months. The cumulative mortality rate in 1347-1352 is thus a good proxy for the monthly mortalityrate that people were facing at the time when a persecution occurred.

Row 1 of Table 1 presents our baseline result. The constant is 0.831*** which reflects the fact that,on average, there was a very high probability of a persecution in 1347-1352. The effect of mortalityon the persecution probability, however, is negative, at -0.009***. This effect is strong, since aone standard deviation increase in mortality (⇡ 0.18) is associated with a 0.34 standard deviationreduction in the likelihood of a persecution. Figure 3(b) depicts this result non-parametrically.The persecution probability increases to 0.8 when mortality reaches 20%, and then decreases to0 as mortality increases to 80%.18 Rows 2-4 then explore the long-run impact of the plague onpersecutions. We find an effect that is half the size of our baseline coefficient in 1353-1400 (row 2)but no effect in the following centuries (rows 3-4).19

4.2. Investigating Causality.

Section 3. suggests that the intensity of the plague was not well explained by characteristics of thetowns affected. We now provide further evidence that the impact of the Black Death was plausiblyexogenous to other factors which may have affected the likelihood of a persecution during theperiod 1347-1352.

A downward bias is more problematic than an upward bias as we then overestimate the“protective” effect of the plague. The effect is biased downward if towns where persecutionswould have occurred anyway during that period were also non-coincidentally affected by lowermortality rates. An upward bias is less of an issue since we then underestimate the protectiveeffect. We discuss below several potential biases and identification strategies.

18One town with a mortality rate of 0% did not persecute its community. However, since we use a bandwidth of 5percentage points of mortality for the local polynomial smooth plot, the mean persecution rate at the origin is 0.4.

19Our finding is entirely consistent with that of Voigtlander and Voth (2012) who find that antisemitic attitudes canpersist over centuries. Our results simply suggest that the direct “protective effect” of Black Death mortality on thelikelihood of a persecution had dissipated several decades after the shock itself.


Correlates of Mortality. In Table 2, we show that mortality rates were uncorrelated with mosttown characteristics that could also have caused persecutions in 1347-1352. We define our townlevel characteristics according to whether they proxy for geography (1), trade and human capital(2) or institutions (3). While several variables are significant in columns (1)-(3), latitude andlongitude are the only variables significantly associated with mortality once we include all thecontrols altogether (see (4), -1.90** and 1.09** respectively). The plague indeed spread from theSouth-East and was initially more virulent due to epidemiological reasons.20

Parallel Trends. Rows 5-8 in Table 1 show that mortality is not associated with persecutionprobability in the period prior to the Black Death, whether we use: 1341-1346 (6 years before1347, as the Black Death lasted 6 years), 1321-1346 (25 years), 1300-1346 or 1200-1346. Paralleltrends suggest fixed effects regressions would give us similar results. Since the Black Death lasted6 years, we create 10 six-year periods t both before 1347 (1297-1302, ..., 1341-1346) and after 1352(1353-1358, ..., 1398-1403). For 172 towns i that had Jews at one point in 1297-1403 and for whichwe know the mortality rate, we regress a dummy if there was a persecution in town i in periodt on mortality in the same town and period, for the town-periods where Jews were present. Weinclude town fixed effects and period fixed effects, and posit mortality is nil outside 1347-1352.When doing so, the baseline effect remains unchanged (see Web Appendix Table A.1).21

Size of Jewish Community. The fact that we capture Jewish presence via a dummy could biasour estimates. If mortality is lower in towns with larger communities and the probability of apersecution is higher in towns with larger communities, simply because they are larger, then thiscreates a downward bias. Conversely, if mortality is higher in towns with larger communities andpersecutions are less likely in such towns because larger communities are better able to defendthemselves against potential persecutions, this also creates a downward bias.

Jews may have been less exposed if they had better hygiene practices or lived in isolated areaswithin towns. Towns with more Jews should then have had lower aggregate mortality rates. Atthe same time, Jewish areas were often overcrowded as they had to live in the few streets reservedfor them. If this was the case, towns with more Jews could have had higher mortality rates.

These biases are unlikely to be significant for the following reasons. First, it is unlikely that Jewsexperienced differential mortality. The plague was mostly bubonic, which limited the role thatcharacteristics of the community could play.22 Second, Jews comprised a small share of the towns,minimizing the effect of any differential mortality between Jews and non-Jews on town mortality.

20The R-squared terms are not exactly equal to 0 when studying the correlation between mortality and trade andhuman capital (see (2), 0.12) or institutions (see (3), 0.15). This is because some of the variables are also correlated withlatitude and longitude. For example, Roman Roads were more dense in the South.

21This result holds even if we drop the 10 post-1352 periods since there were plague reoccurrences after the BlackDeath and consistent data does not exist on the specific mortality rate associated with each reoccurrence.

22For example, in Marseille, the mortality rate of the Jewish population was 50% during the Black Death versus 55%for the whole city (Sibon, 2011). The argument that the Jewish practice of ritual bathing would lead to lower mortalityrates is not well grounded as ritual baths often used stagnant water (see van Straten, 2007, 47) and, contrary to commonmythology, bathing was common among the Christian populations of medieval Europe.


For 30 sample towns for which we have information on the size of their community, the medianpopulation share of Jews was 6.3%. Third, Table 3 shows that results hold if we add proxiesfor the size of the community and the organization of the town (rows 2-6). Row 1 reproducesthe baseline results (-0.009***). Row 2 drops the towns for which we know that Jews accountedfor more than 5% of the population. Row 3 adds dummies for whether the town had a Jewishcemetery/quarter/synagogue circa 1347. Communities with a Jewish cemetery and a synagoguewere larger, whereas towns with a Jewish quarter were segregated. Row 4 controls for the firstyear Jews were present and the last year of reentry in the town before 1347 as towns that had Jewsfor a long time may have had a larger community. Row 5 controls for a Jewish centrality index thatmeasures to what extent the town is surrounded by other towns with a community. This indicateswhether Jews are well-established in the region.23 Row 6 restricts the sample to towns where Jewssettled after 1290, the year Jews were expelled from the Kingdom of England, since their resultingcommunity size might be more exogenous. The effect is weaker as sample size decreases to 36.

Similarly, Web Appendix Table A.3 shows results hold if we focus on towns where the communitywas likely large (small): (i) towns with (without) a Jewish cemetery/quarter/synagogue; (ii)towns where the year of first/last entry is below (above) the median year in the sample; (iii)and towns with a Jewish centrality index above (below) the median in the sample.

In addition, Web Appendix Table A.4 shows that mortality was not correlated with these proxiesfor the size of the community, thus minimizing this concern. Likewise, for 172 cities with mortalitydata and that do not belong to the British Isles or Scandinavia, we find no correlation betweenmortality and whether there was a Jewish community in the town. We exclude these regionsbecause there was a blanket ban on the presence of Jews for their whole territory.

A second, related, question is whether our mortality rates fail to include the deaths due topersecutions, thus mechanically causing an upward bias. However, an upward bias is lessconcerning for us. Moreover, Jews generally comprised a small population share of the towns. Inrow 7, we drop 8 towns for which we know that a significant number of Jews were persecuted.24

Third, towns with more persecutions in the past may have had a smaller community by 1347.Furthermore, towns with a history of persecutions may have had more antisemitic residents,which would then raise the Black Death persecution rate. Conversely, towns with morepersecutions in the past may have been less likely to persecute their community during the BlackDeath if there were fewer Jews left to persecute. In rows 8-11 we control for previous persecutionsusing a dummy indicating any persecution or the number of years with a persecution, whether

23For town i, and other towns j 2 J (363 towns with a Jewish community circa 1347) or j 2 A (all 1,869 towns), theJewish centrality index is equal to

Pj2J D��

ij ÷P

j2A D��ij ⇤ 100 with Dij the travel time between city i and city j. If

all surrounding towns have a Jewish community, it will be close to 100, and 0 otherwise.24We do not know how many Jews were persecuted in other towns. However, the fact that we could not find the

number of victims for these other persecutions suggests that the numbers were not as high as for the 8 towns droppedas persecutions involving more victims should have been better documented than persecutions with fewer victims.


one generation (1321-1346, ⇡25 years) or two generations (1300-1346, ⇡50 years) before. Row 12then shows that results hold if we drop 10 cities for which information on Jewish presence circa1347 was directly inferred from information on the occurrence of a persecution.

Presence of Jewish Community. We focus on towns with a Jewish community since, byconstruction, there cannot be a persecution without a community. However, if Jewish presenceis better measured in towns with high mortality and low persecution, or low mortality and highpersecution, our results may be spurious. In our sample, there is only one town where Jewsentered after 1347. Web Appendix Table A.3 shows that results hold if we drop that city and relyonly on towns whose Jewish presence was pre-determined to the Black Death, i.e. towns: (i) witha Jewish cemetery/quarter/synagogue, (ii) whose year of first entry/last reentry in the town wasbelow the median in the sample, and (iii) whose Jewish centrality index was above the median.

Outliers. In row 13, we drop the towns with the 25% highest and the 25% lowest mortality ratesto ensure that our results are not driven by outliers that may have had high, or low, mortality ratesfor specific reasons. In general, no community was prepared to deal with the Black Death. It wasattributed to the “vengeance of God” or the “‘conjunction of certain stars and planets” (Horrox,ed, 1994, 48-49). Therefore, there was little variation in a town’s ability to deal with the plague.Historians report that some towns had either natural baths (Nuremberg, Baden-Baden) or tried totake action in response to the plague (Venice). Results hold when we drop these towns (row 14).

Controls. A downward bias is possible if any of the characteristics of Table 2 increases (decreases)mortality and decreases (increases) the probability of a persecution. For example, if being ona trade route was positively correlated with mortality and negatively correlated with a town’spropensity to persecute, this would be a source of downwards bias. Alternatively, if being locatedin rugged terrain made a town less susceptible to the plague but more likely to persecute Jews, thiswould be a source of upwards bias. Table 2 has shown, however, that mortality is not significantlycorrelated with most characteristics. Only latitude and longitude have any relationship since theplague came from the South-East and was more virulent initially. In row 15, we include all ourcontrol variables in the regression. However, with only 124 observations, including 27 controlsreduces the degrees of freedom. In addition, as mortality is correlated with latitude and longitude,there is a risk of overcontrolling when adding all controls. The size of our coefficient remainssignificant though smaller in size (-0.005*). When not controlling for latitude and longitude (andtemperature whose correlation with latitude is 0.73), the point estimate is -0.006** (row 16). Notethat we still obtain significantly negative effects when adding dummies for the Holy RomanEmpire, cultural areas, or linguistic areas (see Web Appendix Table A.2).

IV1: Proximity to Messina Market access to Messina should predict plague virulence, since theplague was more virulent initially. We construct an instrumental variable based on a town’s logmarket access to Messina (see Figure 6(a)), conditional on a town’s log market access to all 1,869towns (see Figure 6(b)). Market access to Messina m for town i is defined as MAim = ⌃(Lm÷⌧�im),


with Lm being the population of Messina in 1300, ⌧im the computed travel time between town i

and Messina, and � = 3.8.25 Controlling for market access to all 1,869 towns captures the fact thatsome towns were better connected overall, and may have thus been more likely to receive infectedrats/fleas as a result. In other words, we exploit the fact that it was the specific connectedness toMessina, and not connectedness overall, that mattered for mortality.26

The exclusion restriction is satisfied as long as network proximity to Messina is not correlatedwith factors that directly influenced the persecution probability. Sicily was an urbanized andcommercialized part of Europe but its economy resembled that of its Mediterranean neighbors(Epstein, 1992). Sicily did not treat its Jewish population differently from the rest of Europe(Backman, 1995, 148-150). Messina was not a particularly important city. For example, it wasonly the 45th largest city in our full sample in 1300. There was no reason for distance to Messinato be correlated with Jewish community characteristics. Jews settled in Europe long before beforethe Black Death, and with the exception of the British Isles and Scandinavia, most states hadJewish communities at the onset of the Black Death. It is thus reasonable to believe that networkproximity to Messina should not be directly correlated with persecution probability.

As shown in row 2 of Table 4, the first stage of this instrument is strong (F: 31.0, coeff.: 4.42***).The second stage yields a coefficient estimate that is negative and similar in magnitude to ourbaseline estimate of row 1 (-0.016***, not significantly different from -0.009***). To ensure that ourinstrument is not picking up additional unobservables we control for latitude and longitude andlatitude and longitude squared (row 3). The first stage estimate is weaker (F: 4.3, coeff.: 3.46**) butwe still obtain a negative coefficient that is not significantly different from -0.009*** (Web Appx.Table A.5 shows the full 1st-stage regressions). In row 4, we show the reduced-form effect of theIV — log market access to Messina in 1300 — on persecution probability, but then verify that theIV is unrelated with persecution probability in the preceding generation(s) (1321-1346 and 1300-1346, see rows 5 and 6). In Web Appendix Table A.6, we also verify that this IV is not correlatedwith the log growth rate of town population in 1200-1300, conditional on log market access to alltowns and log initial population in 1200 (to control for mean reversion). In Web Appendix TableA.7, we then show that the IV effects hold if we simultaneously control for all proxies for Jewishcommunity size and past persecutions (cemetery, quarter, synagogue, years of first and last entry,Jewish centrality index, and dummies if persecution in 1321-1346 and 1300-1346).

Despite the fact that we control for market access to all towns, one might worry that market accessto Messina is correlated with market access to other cities, and that better connectedness to these iscorrelated with persecution probability. Web Appendix Table A.7 shows results hold if we controlfor log market access to (i) Genoa and (ii) cities in the Middle-East and North Africa (MENA).Kaffa was a trading colony established by Genoa on the Black Sea (Epstein, 1996, 143). The ships

25Note that we log market access because it is not bounded whereas mortality and persecution probability are.26Note that the correlation between the two market access measures is indeed lower than 1, at 0.51, since Northern

European towns also had access to large towns other than Messina.


carrying the plague from Kaffa were bound to Genoa, and stopped in Messina because sailorswere sick. By controlling for market access to Genoa, we capture the fact that the North-Westof Italy was a wealthy area of Europe then, and we exploit the fact that the ships stopped inMessina and not a different city.27 We control for market access to the MENA using informationon the five largest MENA cities in 1300 according to Chandler (1987). Lastly, a more realisticmeasure of market access would account for variation in local tariffs and currencies. Countingthese is an impossible task as the vast majority of these are not recorded by historians. In anycase, these local tariffs were likely endogenous to local town characteristics, so including them inour analysis could introduce additional bias.28 Instead, Web Appendix Table A.7 shows resultshold if we use an ahistorical measure of market access as our instrument, more precisely the logEuclidean distance to Messina, conditional on average log Euclidean distance to all 1,869 towns.29

IV2: Month of First Infection. A related source of exogenous variation in mortality is timingof first infection, not just because of proximity to Messina, but also because there was a lot ofrandomness in the local patterns of the plague, depending on where infected rats/fleas went.Figure 7 plots mortality rates against the date that the town was first infected (n. of months sinceOctober 1347). Towns infected later, indeed, had lower mortality rates. Using the number ofmonths since October 1347 as an IV, the coefficient is -0.028*** (row 7 of Table 4, F: 33.3, 1st-stagecoeff.: -0.87***). This is precisely estimated, however, this estimate is also significantly larger thanour OLS coefficient (-0.009). In row 8, we also control for latitude and longitude and latitude andlongitude squared in order to rely more on the random component of the spread of the plague.The F-stat decreases to 7.3 (1st-stage coeff.: -0.83***, see Web Appx. Table A.5 for the full 1st-stage regressions), and the coefficient is less precisely estimated (not significantly different from-0.009***). In row 9, we show the reduced-form effect of the IV on persecution probability, butthen verify that the IV is unrelated with persecution probability in the preceding generation (s)(1321-1346 and 1300-1346, see rows 10 and 11). Web Appendix Table A.6 shows that this IV is notcorrelated with town population growth in 1200-1300. We combine IV1 and IV2 in row 12. Theeffect is -0.019* (F-stat of 4.4, not significantly different from -0.009).

Plague Reoccurrences. The plague reoccurred intermittently over the next centuries. Using datafrom Schmid et al. (2015) based on Biraben (1975), we study the relationship between plaguereoccurrence and the persecution of Jews. We focus on 415 towns with Jews at one point in 1353-1598. Since the Black Death lasted 6 years, we create 41 six-year periods t (1353-1358, . . . , 1594-1598). We then regress a dummy if there was a persecution in town i in period t on the numberof years with a plague outbreak within 5 km, or 100 km, from the town during the period, for thetown-periods where Jews were present. We include town fixed effects and period fixed effects.Note that we also use outbreaks within 100 km because Biraben recorded plague outbreaks forlarger cities only (Roosen and Curtis, 2018), hence the need to rely on local buffers around each

27The correlation between log market access to Messina and log market access to Genoa is lower than 1, at 0.69.28We also use � = 3.8 (from Donaldson (2018)), which implies a high cost of distance on trade.29Results also hold if market access to all towns exclude Messina (Web Appendix Table A.7).


city. There is also no data on the mortality rate associated with each outbreak, so our measureis very imperfect. Nonetheless, Web Appendix Table A.8 shows there is a negative effect onplague reoccurrences on persecution probability (-0.005*** when using 100 km; -0.006 (p-value:0.23) when using 5 km).30

Distribution of Rats. The Black Death was transmitted by the fleas of the black rat (Rattusrattus). However, their populations began to decline after the introduction of the brown rat (Rattusnorvegicus) in the 18th century (Christakos et al., 2005), which coincided with the end of plaguereoccurrences in Europe. Note that little is known about the geographical distribution of Black ratsat the time of the Black Death, and we cannot infer it from the distribution of brown rats today(for which little data also exist). We thus do not use rats for identification purposes.

Summary. Our analysis of town characteristics, parallel trends, Jewish community size, outliers,controls, and IV strategies suggest there is a plausibly causal negative effect of mortality onpersecution probability. The identification strategies suggest effects that range between about -0.006 and -0.019, but are not significantly different from our baseline OLS effect of -0.009. In therest of the analysis we will thus employ the intermediary OLS estimates as our baseline.

4.3. Robustness

Preventive Persecutions. Historians note that some communities heard accusations that the Jewswere poisoning the wells and killed their Jews prior any outbreak of plague (Ziegler, 1969, 103).These persecutions raise several problems for our analysis. First, our hypothesis is that peopleand authorities observe how many other people are dying and then decide to persecute theircommunity. If most of the persecutions in our sample were preventive persecutions, our resultswould be coincidental. Second, preventive persecutions were likely to occur in the later years ofthe Black Death (1347-1352). But places infected later also had lower mortality rates.

Table 5 considers these issues. Baseline results are shown in row 1. We first add a dummy fortowns which were first hit by the plague after September 1348 as they are the towns liable topreemptively persecute their Jews (row 2). Indeed, September 1348 is the month when Jews weretortured at the Castle of Chillon in Switzerland and forced to admit to poisoning wells which, inturn, started the rumor that Jews had caused the disease (Horrox, ed, 1994, 68). In the analysisbelow the 10 possible preventive persecutions in our sample all took place after September 1348.

Second, we control for the log distance to Chillon and its interaction with the post-September1348 dummy in order to control for the effects of the diffusion of the rumor from Chillon after thatmonth (row 3). Next, we add a control for the log distance to the first ten towns to be warnedby letter of a Jewish conspiracy and its interaction with the post-September 1348 dummy in orderto control for the effects of the diffusion of the rumor (row 4). We then include a control for thelog distance to the Rhine and its interaction with the post-September 1348 dummy as the rumorsalso diffused along the Rhine (row 5). Lastly, in row 6 we control for the log distance to the path

30However, given the limitations of this analysis, we do not pursue it further.


of the Flagellant movement and its interaction with the post-September 1348 dummy as they arealleged to have massacred Jews in their attempts to ward off the plague (Nohl, 1924). None ofthese factors significantly affect our estimates (which decrease but retain their precision).31

We also drop towns where we suspect persecutions may have preceded the plague itself. In rows7 and 8 we drop the towns where it is likely or possible that a persecution preceded the plaguebased on our knowledge of the year that the persecutions occurred.32 In rows 9 and 10 we droptowns where data on the year and month of first infection and year and month of persecutionsuggest that these persecutions were likely to have been preemptive or were possibly preemptive.

Alternative Outcomes. In Table 6, we explore whether there were differences in the intensity ofpersecution. Our results are stronger for pogroms than for expulsions (-0.007*** in row 2 vs. -0.004** in row 3, significantly different at 5%). The fact that the effect is larger for pogroms mayreflect the fact that pogroms involved the killing of Jews. Expulsions, by contrast, were moreorderly events and involved the Jews being expelled from a city into the nearby countryside fromwhich they could return. In this case, if a higher mortality rate increases the economic value ofJews, a town may prefer to expel its community instead of killing its members.

Among the 58 towns with a persecution we investigate in row 4 if the persecution led to Jewsbeing expelled or all Jews being killed (“Annihilation”). We find a negative effect of mortality onthe probability of an exit. This implies that if the mortality rate is high the populace and/or theauthorities prefer to kill a few Jews rather than losing its entire Jewish community. This is in linewith our baseline result that towns persecute Jews “less” at higher mortality rates.

Among the 45 towns with a pogrom, we study if the pogrom was particularly violent, for examplewhether the community was annihilated (row 5), whether some Jews were burned alive (row 6),or whether the pogrom was perpetrated by a mob (row 7). We find that a higher mortality rateled to a lower “intensity” of the pogrom, as it reduces the likelihood of an annihilation (row 5,-0.008*) and a mob being involved (row 7, -0.010***). We find a negative, but not significant, effectfor burning. Lastly, pogroms were “less” violent when mortality was high (row 8, -0.010**).

Lastly, we know if the local authorities attempted to prevent a persecution. Row 9 shows resultshold if we consider as a persecution any persecution that was successful or unsuccessful. Row10 shows results hold if we consider as a persecution only the persecutions where no preventionattempt was made. In row 11, among the 61 towns with a successful or unsuccessful prevention,we show that there is a positive but not significant correlation between a dummy equal to one ifthere has been any attempt to prevent the persecution and mortality.

31Results hold if we use network distances instead of Euclidean distances (results available upon request).32In four towns, persecutions took place in 1349, but the Black Death did not hit these before 1350. It is possible that

these were not preventive persecutions if the year of infection was measured with error. In four towns infected in 1350,persecutions took place in either 1349 or 1350, although we are not sure of the year due to conflicting sources.


External Validity. There could be concerns that sample size is small. In row 2 of Table 7, we use asregression weights population in 1300. By doing so, less weight is placed upon small towns. Theeffect becomes stronger. Given that our 124 towns are capturing 66.7% of the total population ofthe 363 towns with a Jewish community, having more towns would not change results (unless theyare very large). We also use several albeit imperfect methods to obtain mortality estimates for moretowns. First, the effect remains similar if we use the mortality rate of the closest neighboring townwith mortality data if this town is within 100 km (row 2, N = 307). Second, we create estimates ofmortality based on spatial extrapolation.33 In row 3, we use these extrapolated estimates to showthat our analysis holds for the full sample of 363 communities. Third, Christakos et al. (2005)report mortality rates for selected provinces (e.g., “Languedoc”). When we attribute to each townwhose mortality is missing the rate of its province, the effect is similar (row 4, N = 159).34

Since the 124 towns represent 66.7% of the total population of the 363 towns with a Jewishcommunity, one could object that our sample is biased towards large towns among the groupof towns with a community. Kernel distributions of 1300 populations and Kolmogorov-Smirnovtests confirm that our 124 towns are, on average, larger than the 363 towns. If complementarities,scapegoating and other effects vary with population size, our estimated effect may differ from theaverage effect. We thus show results hold if we: (i) control for log population in 1300 (row 5); (ii)drop the towns in the top and bottom 25% in population size (row 6); and (iii) drop the townsin the top 10% in population size (row 7), as doing so is enough to ensure that our sample is notsignificantly different in size from the sample of 363 towns (now shown).

Measurement Concerns. In row 8 we report our estimates using a sample based solely on theEncyclopedia Judaica. In row 9 we drop 12 towns where we cannot be entirely certain that therewas a community intact in 1347 and 20 towns where we cannot be entirely certain about thefate of the community during the period.35 We drop (i) mortality estimates based on literarydescriptions (row 11); (ii) estimates based on desertion rates (row 12); (iii) estimates based onnumerical estimates (row 13). In row 14 we only use numerical based estimates. Another approachis to focus on those cities that are either in the bottom 25% of least affected cities or in the top25% of most affected cities, since measurement errors are more likely when comparing townswith mortality rates around 50% (row 15). In row 3 we analyze all 263 cities for which we havemortality estimates. We implement this test because it is possible that some of the towns that weclassify as not having Jews actually had Jews in 1347. Results hold overall.

33We create a two-dimensional surface of predicted plague mortality using an inverse distance weighted function ofknown mortality rates for the full sample of 263 towns with mortality data. For every point on the surface a predictedmortality rate is then generated using the closest 15 cities within an approximately 1,000 km radius circle around thepoint. Details can be found in Web Appx. Section 5 (map of extrapolated mortality rates shown in Web Appx. Fig. A.3).

34One problem here is that several towns now use the same provincial estimate (e.g., Aragon (7 towns) and Bohemia(5)). Our results are then robust to dropping towns located within different modern country borders (Web Appx. TableA.9): France, Germany, Italy, Portugal and Spain. Other countries only contribute 7 towns or less.

35The 12 towns include 8 French towns for which information is sparse following the expulsions of 1306-1322, and3 Tuscan towns in which Jews settled in the 14th century but for which we cannot be sure of the year. The 20 townsinclude towns for which sources mentioned a persecution, but without providing corroborating details about it.


Another related concern are non-classical measurement errors. If the community is small, andmortality high, there may not be any Jew left to persecute. We then misclassify as having Jewscities that did not have Jews when deciding to persecute, particularly so for high-mortality cities.This creates a negative relationship between mortality and persecution probability. However,while Jews accounted for a small share of the population of the towns (mean of 7.5%, for the30 towns for which we have such data), towns were not that small on average (mean of 18,000inhabitants for the 124 towns), so the size of their community was not negligible (mean of 1,060inhabitants for the 30 towns). Even with an extreme mortality rate, there must have been Jewsleft to persecute. Second, the mean duration of the plague in each town was 5 months, and notall people died at once, so there must have been Jews alive the first months. Third, Figure 3(b)showed that the persecution probability almost linearly decreases for mortality rates higher than20%. Results should not be driven by high-mortality outliers. Row 7 of Table 7 confirms this bydropping those cities in the top 25% of most affected cities (mortality above 50%).36 Lastly, in row15 we employ Conley standard errors (threshold of 100 km) to account for spatial auto-correlation.

One could imagine that at high mortality rates there is no one left to report a persecution, so thatpersecution rates are underestimated. However, the Encyclopedia Judaica does not only rely ondocuments provided by communities themselves, but also official city records, personal diaries,cemetery records, etc. Second, most towns were sufficiently large initially, and their estimatedresidual population in the immediate wake of the Black Death was also large (mean of 11,000inhabitants). Even in the top 25% of most affected cities, the residual population remainedlarge (mean of 6,500 inhabitants). Third, persecution probability almost linearly decreases withmortality and results hold when dropping or keeping the top 25% of most affected cities.

5. Discussion of the Mechanisms

One approach to explore the mechanisms is to compare towns with different mortality rates, andobserve how the de facto preference for diversity and the economic valuation of having Jewsaround differ across these towns. Obviously, no data exist on the level of antisemitism, theeconomic value of the services provided by Jews, and the cost of committing a persecution foreach town during the months in which the town was infected. We only observe actions. However,since actions reveal preferences, we can use our conceptual framework, our econometric results,and qualitative evidence, to discuss which patterns of the scapegoating and complementaritieseffects could make sense in our context. We follow this approach in Section 5.1. below.

Another approach is to compare towns with similar mortality rates but different characteristics thatmay affect the relative strength of the scapegoating and complementarities effects. This approach,which we follow in Subsection 5.2., allows us to quantify the importance of the effects.

36Web Appendix Table A.3 shows results hold if we restrict the sample to towns where the community was likely tobe large enough for the plague not to annihilate the community, i.e. towns with a Jewish synagogue/quarter/cemeteryor towns where Jews have been living for a long time.


5.1. Qualitative Evidence on the Mechanisms

To explain the patterns in the data, a plausible explanation is that: (i) at low levels of mortality, thescapegoating effect is strong, so that the persecution probability is high (relative to the pre-plaguecase); and (ii) at high levels of mortality, the scapegoating effect decreases and/or is offset by acomplementarities effect that reduces the likelihood of a persecution when mortality increases.

Scapegoating: Blaming Jews. The Black Death persecutions provide a classic example ofideological scapegoating. The Jews were blamed for spreading the disease (Cohn, 2007; Aberth,2009). In particular, the chroniclers accounts suggest that it was the sheer scale and devastatingimpact of the Black Death that made contemporaries feel that it was either the wrath of God or partof a grand conspiracy and the responsibility of the antichrist or his followers (Nohl, 1924; Moore,1987; Horrox, ed, 1994). In line with the scapegoating theory, this suggests that the probability ofpersecuting the Jews was likely to increase with plague virulence (Glick, 2002, 2005).

Scapegoating: Updating Priors. An alternative explanation is that the scapegoating effect wasinitially increasing in mortality but as mortality increased further it declined as non-Jews revisedtheir prior that Jews were to blame. For example, in September 1348, Pope Clement VI issued aBull contradicting the libel against the Jews (Chazan, 2010, 153-154), observing that “it cannot betrue that the Jews [. . . ] are the cause or occasion of the plague, because throughout many partsof the world the same plague, by the hidden judgement of God, has afflicted and afflicts the Jewsthemselves” (Horrox, ed, 1994, 222).

However, by then, 16 persecutions had already taken place in our data, including a fewpersecutions around Avignon. The Papal Bull also did not prevent the 42 persecutions in ourdata after September 1348. One reason for this was that the authority of Papacy had beengreatly weakened by the relocation to Avignon in 1309 and the perception that the Pope wasthe “puppet” of the French king (Mollat, 1963; Renouard, 1970). Bishops and Archbishops hada lot of independent discretionary authority due the limitations of communication and transporttechnologies. The low clergy, moreover, often held more antisemitic views (Cohen, 1982). It istherefore unlikely that the Bull had any effect beyond Avignon. Also, remember that our effectswere unchanged when controlling for whether the town had a Jewish quarter or not (row 3 ofTable 3). Furthermore, the theory of “motivated reasoning” suggests that once individuals havedecided to blame a group for a misfortune, they are resistent to countervailing evidence (Kunda,1990). The results are thus unlikely to be driven by non-Jews updating their priors.

Scapegoating: Cost of Committing a Persecution. We could also imagine that Christiantownsfolks were less able to organize themselves to persecute the Jews at higher mortality rates.Indeed, if many non-Jews were ill (or simply dead), or had to take care of the sick, they may havenot been able to commit a persecution. Note that most persecutions in our sample were pogroms.Expulsions were often orderly events, and thus required more organization than pogroms. On thecontrary, and unlike the Holocaust, pogroms required almost no organization.


First, Jews were easily identifiable, often lived in specific areas and, more importantly, wereforbidden to bear arms (see Baron, 1967). Second, records indicate that persecutions were oftencarried out by spontaneous mobs of villagers and townsfolks.37 In general, these villagers andtownsfolks were local people armed only with axes, knives, pitchforks or wooden sticks (Aberth,2009). Third, the plague lasted on average 5 months in each town. The period was thus longenough for people to organize themselves to commit these persecutions, especially as most ofthem took place over one or two days only. People were also not dying all at once.

In addition, if organizational costs were high enough to prevent persecutions at high mortalityrates, this should especially affect our results when comparing places with low and high mortalityrates. However, we showed that the protective effect was stronger when dropping the towns withthe 25% highest and the 25% lowest mortality rates (i.e. mortality rates below 25% or above 50%),so places with presumably low and high organizational costs (row 13 of Table 3). Likewise, ourresults hold when dropping only the 25% highest mortality rates (i.e., mortality rates above 50%),since the high organizational costs should affect high-mortality towns (row 12 of Table 6). Witha mortality rate of 50%, as long as the community has at least two residents, there should be oneJew left to persecute. However, in our sample of 29 towns for which we know the size of thecommunity, the mean Jewish population is 1,060, and the five smallest communities have 22, 57,128, 132 and 150 individuals. It is thus very unlikely that no Jews were left to persecute.

Additionally, a high mortality rate is only a problem if there are no non-Jews left to commit thepersecution. We show that results hold if we restrict our sample to towns with a population in1300 below or above the median (rows 13-14 of Table 6). Likewise, results hold if we restrict oursample to towns with a low or high population in the immediate aftermath of the plague. Moreprecisely, we estimate this “residual” population as the population of the town in 1300 multipliedby (100 - the mortality rate of the town)/100, and then show results hold if we only use towns witha residual population below or above the median (rows 15-16 of Table 6). Similarly, it is also notthe case that a higher mortality rate would reduce the chances of there being a Jewish communityat all, or anyone at all to report a persecution.

Complementarities. Evidence suggests that towns internalized the economic benefits brought byJews. Rulers frequently “anticipated Jewish contribution to the economy” (Chazan, 2010, 102).38

Even during the Black Death period, historians note that the “economic importance” of certaincommunities enabled them to weather persecutory storms.39 Jews were also frequently given tax

37In Ulm we are told that on Jan. 30, 1349, a mob “stormed” the Jewish quarter; in Magdeburg, the “citizens andpeasants of the vicinity fell upon the Judendorf, pillaged it, and burned many Jews in their houses”.

38In the 11th century the Bishop of Speyer argued that “the glory of our town would be augmented a thousandfoldif I were to bring Jews” (quoted in Chazan, 2010, 101). Similarly, peasants in Tuscany petitioned for the admission ofJewish lenders to make credit more abundant (Botticini, 2000).

39This is the case of Regensburg where Jews conducted “Moneylending to secular and ecclesiastic princes, in additionto the merchants of the city, was conducted on a large scale” and which survived the “Black Death persecutions(1348/49) and to reach new heights of prosperity by offering asylum to rich refugees” (Wasserman, 2007, 188).


breaks and other encouragements to resettle in the aftermath of a pogrom.40 More generally, theimportant role Jews played as moneylenders is documented by numerous historians (Roth, 1961;Baron, 1967; Botticini and Eckstein, 2012) and economists Pascali (2016); Becker and Pascali (2016).

This qualitative evidence is consistent with a positive complementarities effect whereby Jews aremore than proportionally needed when mortality increases. An alternative pattern, in which theJewish contribution to economic activity is valued less as mortality increases, is possible. Wecould imagine that as a large number of people die suddenly, people and rulers become morepresent-biased and internalize less the long-term externality of having Jews around. In this casethe incentive to persecute the Jews increases. However, historical evidence suggests that thisexternality was internalized. In addition, given that Jews were likely to have been more thanproportionally blamed when mortality increased, the lower persecution probability observedat higher mortality rates can best be explained by a complementarities effect increasing withmortality. In other words, the incentive to persecute must have been dominated by the incentiveto protect them at higher mortality rates.41

5.2. Quantitative Evidence on the Mechanisms

To study these channels quantitatively, we interact mortality with various pre-Black Death towncharacteristics and look at how the effects of mortality on the probability of persecution wereaccentuated or attenuated depending on these. We estimate a series of regressions based on:

Pi,1347�52 = ↵+ �Mi,1347�52 + �Mi,1347�52 ⇤Xi + �Xi + ✏i , (4)

where Pi,1347�52 is a dummy equal to one if there is a persecution in town i in 1347-1352, Mi,1347�52

is the cumulative mortality rate (%) in town i over the period 1347-1352, and Xi is a dummyvariable proxying for a pre-plague town characteristic. � is our main coefficient of interest andcaptures whether the protective effect of higher mortality was accentuated or attenuated in townswith this characteristic. As we control for the mortality rate, we are comparing towns with thesame mortality rate, but different pre-shock characteristics. Note that we are also controlling forthe individual effect of the town characteristic on the persecution dummy.

Tables 8 and 9 report the effect of mortality (�), the effect of its interaction with the dummyvariable proxying for the characteristic (�), and the combination of these effects (� + �), to seeif the interacted effect is strong enough to offset the protective effect of mortality. Note that thedummies that start as “Close to . . . ” are equal to one for the towns that are in the bottom 10%of the distance to the described location. In addition, the interacted effect of mortality and the

40Jews returned to Nuremberg soon after the plague persecutions (Berenbaum and Skolnik, 2007b). In Aragon,where the impact of the plague was severe, King Pedro strove to protect the Jews and in the wake of the plague“was determined to restore the Jewish aljamas to a healthy economic state”. He barred creditors from bringing lawsuitsagainst Jews or collecting debts against them for a year while individuals who moved into settle land left empty due tothe mortality associated with the plague were forced to inherit the debts owed to the Jews (Shirk, 1981, 363).

41Many authorities such as in Cologne and Zurich tried to protect their communities because of their economic value.


dummy is not necessarily causal, as the characteristic may be correlated with other characteristics.

5.3. Evidence on Scapegoating

In Table 8 we examine the effects of scapegoating on the probability of persecution.

5.3.1. The Well Poisoning Libel

The Origin and Spread of the Libel. Figure 8(a) shows Chillon Castle, the place of origin of thewell-poisoning libel, and the Rhine river, along which the rumor of a conspiracy spread. Proximityto Chillon (row 1, 0.016*) wipes out the protective effect of high mortality. The effects of proximityto the 10 towns that were first warned of the alleged conspiracy also attenuate any protectiveeffects and increase the probability of a persecution when mortality was high (row 2, 0.017*).Proximity to the Rhine river also reduces the protective effective of mortality (rows 3, 0.017**).42

Attempts to Counteract the Libel. We explore the extent to which the Pope was able to counteractthe libel. We find that close to Avignon there was no protective effect of mortality (row 4, 0.018***).This likely reflects the fact that in many cases the plague arrived and Jews were massacred priorto the Papal Bull. Looking at cities that were the seats of bishops or archbishops, we find someevidence that the scapegoating effect was weakened, but not significantly so (row 5, -0.007). Thisweak effect is consistent with the limited influence that the Pope had on local decisions.

Jewish Community Characteristics. We find that in towns where Jews were recent migrants(less than five years), the protective effect of higher mortality was attenuated (row 6,0.015***). Speculatively, this may be because people made a direct association between therecent arrival of Jews and the plague.43 We consider whether the presence of a Jewishcemetery/quarter/synagogue had an effect on persecutions. The presence of a Jewishcemetery/synagogue suggests that the community was well established. In that case, we wouldexpect the protective effect to be accentuated if it is more socially integrated. The protectiveeffect would be attenuated if non-Jews resent the fact that Jews have their own structures. Theexistence of a Jewish quarter meant that Jews were separated from Christians. This may haveprevented non-Jews from seeing that Jews were also dying. However, it might also have reflectedthe larger size of a community. We find, however, no effects for these community characteristics(row 7). We also find no differential effect for each characteristic considered individually, evenwhen controlling for community size via several proxies (years of first and last entry, centralityindex, dummies if persecution in 1321-1346 and 1300-1346) (see Web Appendix Table A.10).44.

42Historians have traditionally held the flagellants responsible for massacring Jews. Yet, we find no evidence thatthe path of the flagellant movement was associated with the mortality-persecution relationship (see Web Appx. TableA.10). We find, however, that the protective effect was accentuated very close to Narbonne (i.e., for towns in the bottom5% of the Euclidean distance to it). There, “beggars and mendicants”, and not Jews, were accused of having poisonedthe wells, which could explain the lower effect of mortality on persecution probability (Web Appx. Table A.10).

43We do not find the same effects when considering older entries (see Web Appendix Table A.10).44There is no existing data on the lay-out of towns back then, so we do not know the location of the quarter relative

to the main areas of the town then. The only other characteristic for which we have data is walled density (N = 56), but


Ashkenazi Communities. While all Jewish communities share a common genetic legacy thatcan be traced back to the Levant, there are differences between communities. Ashkenazi Jewsshare a number of genetic markers and were characterized by a low rate of admixture for thepast 80 generations suggesting that they remained a genetically distinct population (Behar et al.,2010). In contrast, Askhenazi (as well as Sephardic) Jews share more genetic markers in commonwith southern European populations than they do with northern Europeans (Seldin et al., 2006).Therefore, Ashkenazi Jewish communities were likely to have been culturally more distant fromtheir Christian neighbors than was the case for Sephardic Jews in southern Europe. Looking atconflict in the modern period, Arbath et al. (2015) find that genetic diversity within a countrypredicts violence and conflict. Consistent with this in row 8, we find that mortality conferred lessof a protective effect on Ashkenazi communities than it did on Sephardic communities (0.015**).

5.3.2. Latent Antisemitism.

Past Pogroms. When we include an interaction with a dummy for if there had been a persecutionin the 6 years (1340-1346) prior to the Black Death period (which also lasted 6 years), we finda positive and significant effect (row 9, 0.039*). As we report in Web Appendix Table A.10, thiseffect is weaker if the past persecution dummy is defined for a past period of 25 years in 1321-1346(0.012, only significant at 15%), and nil if it is defined for a past period of about 50 years in 1300-1346. This suggests that a recent history of persecutions, i.e. of antisemitic violence committedby ones’ parents, made a difference and diminished the “protective effect”. However, one couldalso argue that recent persecutions made the community smaller at the time of the Black Death, sothat the stronger protective effect actually reflects the reduced ability of the community to protectitself. That is why we now turn to more historical measures of antisemitism.

Crusades. The First Crusade (1096) saw a marked intensification of antisemitism (Cohen, 1957,77). The call for a crusade to recapture Jerusalem from the Muslims elicited tremendous popularreligious passion and anger against the enemies of Christ who included both the Muslims and theJews (Eidelberg, 1977; Chazan, 1987). The first group of crusaders gathered in France and marchedthrough Germany on their way to Jerusalem and on their way many crusaders attacked Jewishcommunities (see Figure 8(a) for the locations of these pogroms).45 We find that in towns close toa First Crusade-era pogrom, the protective effect of mortality is wiped out (row 10, 0.016**).46

Ritual Murder. The rise of virulent antisemitic tropes over time led to an intensification ofanimosity in the middle ages (Trachtenberg, 1943; Rubin, 2004). Blood libel accusations beganin England in the 12th century and spread through Europe (Stacey, 1998). The libel involved thestory that an organized group of Jews had murdered Christian children in order to obtain blood

we find no differential effect across towns above and below median density (see Web Appendix Table A.10).45Schwarzfuchs (2007, 311) writes: “The sight of the wealthy Rhenish communities acted as an incentive to the

crusaders, who decided to punish ‘the murderers of Christ’ wherever they passed”.46We also collected data on the pogroms that took place during the Crusades of 1147 and 1189 but the later Crusades

were not characterized by large-scale antisemitic violence (see Web Appendix Table A.10).


for the Passover (Ben-Sasson et al., 2007, 74). The myth of the blood libel was spread by preachersand reinforced when Jews were tortured into confessing to the alleged crime. Figure 8(a) showsthe locations of ritual murder charges in the 13th and early 14th centuries. When we interact themortality rate with a dummy for towns close to the alleged location of a ritual murder in the 13thcentury, the protective effect of higher mortality is attenuated (row 11, 0.016**).

Host Desecration. As the belief that the consecrated eucharist contained the body of Christbecame more important in the 12th and 13th centuries, stories of Jews desecrating the host beganto circulate in Europe (Rubin, 1992, 2004).47 By desecrating the host, the Jews were imagined to bereenacting the suffering of Christ. Such charges often inspired acts of antisemitic violence. Figure8(a) shows the locations of host desecration charges in the 13th and early 14th centuries. We findthat accusations of host desecration in the first half of the 14th century (but not earlier) eliminateany protective effect of mortality (rows 12, 0.011**).48

5.3.3. The Liturgical Calendar.

Advent. Advent, the period before Christmas, is the beginning of the liturgical year. During thatperiod, Christians are encouraged to do penance to prepare for the coming of Christ. Consistentwith these motivations, in towns where December was the month of first infection, we find thatthe protective effect of Black Death mortality was reinforced (row 13, -0.022**).

Christmastide. In contrast, the period from Christmas Day (December 25th), when the birthof Jesus is commemorated, until the Feast of the Presentation of the Lord (February 2nd) wasChristmastide. According Matthew’s Gospel (2:1-23), after the birth of Jesus, magi visited Herodthe Great, the King of Jews, to inquire about the location of the “one having been born king ofthe Jews”. Herod, who felt threatened by this newborn, ordered the death of all male babies,an event which came to be known as the Massacre of the Innocents. The commemoration of themassacre was historically connected with the Feast of the Epiphany, on January 6th. During theFeast, Christians celebrated the martyrdom of the Innocents (Archer, 1984, 47).49

Jews were historically viewed as “Christ-killers”. The charge of deicide was at the heart of muchantisemitism up until modern times (Freudmann, 1994; Friedlander, 1997; Donaldson, 2010). Thismade Christmas a period of heightened tensions between Christians and Jews. Jews sometimeshad to pay special taxes on Christmas Day.50 Consistent with this, in towns where January was

47Stacey (1998, 13) writes: “we begin to see Jews taking on a new role, as enemies not only of the body of Christ onthe cross and of the body of Christ in the Church, but also as the enemies of the body of Christ in the eucharist. This isthe notion that lies behind the host desecration charge—that Jews would torture [. . . ] consecrated eucharistic hosts”.

48The non-significant results for ritual murders in the first half of the 14th century and for host desecration in the13th century are shown in Web Appendix Table A.10.

49Archer observes that “At a period in which the individual child was not looked upon with particular tenderness,parental sentiment found a communal channel in the idolization of the Christ child and of the Innocents, the baby boysbutchered in the first attempt to kill Jesus by Herod, the Jewish King” (Archer, 1984, 47).

50This was the case in Trier where Jews gave six pounds of pepper and silks for his clothing to the archbishop andtwo pounds of pepper to the chamberlain. In Ottingen, each Jew over the age of twelve had to pay a gulden to the


the month of infection, higher mortality conferred no protective effect (row 14, 0.016**).

Lent. Lent, which usually starts in February or the first week of March (February 16th in 1348),was a 40-day period of restraint when Christians are supposed to fast and do penance. Consistentwith this, in towns where February or March was the month of first infection, higher mortalityconferred a stronger than usual protective effect (row 15, -0.014***).

Easter. Another period with heightened antisemitism was the period between Easter, in Aprilaccording to the Julian calendar, and the Ascension of Christ (40 days later). Easter celebrationssuch as the reenactments of the passion of Christ made the alleged role of the Jews in the death ofChrist more salient. During Easter Christ was expected to rise and to triumph over his enemies.51

This was reflected in local institutions. Jews were prohibited from going forth in public on thethree days before Easter (see Chanes, 2004, 86). In some cities Jews paid a special tax on Easterto atone for the crime of killing Christ (e.g. Gottheil and Kahn, 1906). This institution sometimesinvolved public humiliation.52 As such, there were many cases of pogroms occurring duringEaster.53 Freudmann (1994) notes that “the annual paschal pogrom became a Christian rite ofspring” (Freudmann, 1994, 300). Consistent with this, in towns where April or May was the monthof first infection, higher morality conferred no protective effect (row 16, 0.011*).54

Overall, these results are consistent with the scapegoating effect.

Agricultural Calendar. An alternative interpretation of some of our seasonal results is based onthe agricultural calendar as April and October were the main months of planting and April-Maywere the lean months when farmers had to borrow money from Jewish moneylenders to buy newseeds and sometimes food that would last them until the next harvest (July).55 These farmers had

emperor on Christmas day (Adler and Singer, eds, 1906).51A favorite Easter sermon from St. Augustine provided a fictitious antisemitic account of Jesus’s death: “The Jews

held him: the Jews insulted him; the Jews bound him; they crowned him with thorns, dishonored him by spitting onhim; they scourged him; they heaped abuse on him . . . ” (Freudmann, 1994, 300). Just as hostility to Jews at Christmas-time manifested itself with the association of Jews with Herod, so at Easter they were closely associated with Judas.

52In Toulouse there was a tradition whereby the Jewish community had to choose a member of the community everyyear to be publicly slapped in the face on Good Friday. In the middle ages this ritual was waived on the condition thatthe Jews pay a special fee to the town (Blumenkranz et al., 2007).

53The tensions around the month of April were further heightened by the fact that Passover typically coincides withEaster and Jews were accused of ritually murdering and sacrificing Christian children as part of this festival (Rubin,2004). Numerous studies suggest that this was a factor during the Black Death. The Jewish community of Toulon weremassacred on 12-13 April 1348, at the same time as the plague reached the city (Cremieux, 1930).

54We investigate the effect for each month individually, by regressing the persecution dummy on the mortality rate,12 month of infection dummies, and the 12 interactions of these dummies with the mortality rate. We then test whetherthe effect of each month is significantly different from our baseline effect of -0.009. Web Appendix Figure A.6 confirmsthat the protective effect of mortality was accentuated in February, March and December, and attenuated in January,May (significant at 15%) and October. There is no clear effect for April now (Easter took place in late April in 1348).

55In ancien regime France, the passage of time was defined by “agro-liturgical calendars” (Moriceau, 2010). Thesecalendars provide information about the agricultural cycles of the medieval period. The agricultural cycle depends onwhether one considers a town in the North or the South of Europe. In our sample, the mean latitude and longitude are45.7 and 6.0, close to Chambery. In France, December and February were “idle” months, January was the month fieldshad to be plowed for the spring sowing, March-April and October were the months of planting, April-May were the


to repay their loan six months later around October and sometimes extending an existing loanfrom the same Jewish moneylenders to buy seeds for the winter planting season.56 The smallerprotective effect of higher mortality in April-May (row 16, 0.011*) and October (row 17, 0.011***)could be consistent with the desire to not have to repay these loans.

5.4. Evidence on Complementarities

In Table 9 we examine the effects of complementarities on the probability of persecution.

5.4.1. Financial Competition and Moneylending.

Financial Competition. We construct a measure of access to financial markets based on distanceto major financial centers at the time: Cahors, Florence, Genoa, Milan, Sienna and Venice (Figure8(b) depicts these cities).57 For cities close to a financial center, the protective effect was attenuatedwhen mortality was high (row 1, 0.008***). In regions with Christian moneylenders, Jews weresubstitutes and hence more vulnerable.

Moneylending. Next, we collect data on whether Jews are mentioned as moneylenders in theEncyclopedia Judaicia. Moneylending is mentioned for 60 towns in our sample (Figure 8(b) depictsthese towns). Where Jews are mentioned as lending money, we find that when mortality is high,the protective effect is reinforced (row 2, -0.009**). In towns where Jews were moneylenders, theymay have become more valuable when the plague was more destructive.58

We find no relationship between plague intensity and persecution probability and other economiccharacteristics of Jewish communities such as whether Jews worked as craftsmen, traders or asdoctors, or are reported as having paid special taxes to the ruler (Web Appendix Table A.10).59

5.4.2. Trade and Urbanization

Log City Size and Log Market Access. Mean mortality was 39%. Many towns, however, lost aneven higher share of their population and their economies collapsed. These towns—and especiallytown authorities—were concerned about their demographic and economic survival in the wake

“lean” months, and July and October were the months of grain and grape harvesting.56Ramirez (2010, 64) writes for the region of Savoy in which Chambery is located (translated): “Most of the loans

[given by Jews] take place in the Spring and the Fall [. . . ] during the planting season and during the lean season.”57Florence was the most important financial center in 14th century Europe (de Roover, 1963). Genoa, Milan, Sienna

and Venice were also important financial and banking centers (Mueller, 1997). Cahors in southern France was a cityknown for its Christian moneylenders (Noonan, 1957). Cahorsins are explicitly mentioned alongside “Lombards” asdirect substitutes for Jewish lenders (see, e.g. the entry for “Switzerland”’ in the Encyclopedia Judaica).

58Nevertheless, as we showed above, conditional on Jews lending money to non-Jews, they were more likely to bepersecuted when the arrival of the plague coincided with the period when their debts were due to be repaid.

59This is unsurprising as the literature agrees that the importance of Jewish merchants had waned by the 14th century.Roth (1961) argues that guilds pushed Jews out of trade and manufacturing into moneylending. However, Botticini andEckstein (2012, 238) argue that guilds cannot explain the shift towards moneylending because by the time they becamepowerful, “the Jews in western Europe had entered and then become specialized and prominent in moneylending forat least two centuries”. Instead Botticini and Eckstein (2012) argue that Jews shifted away from trade to due to theircomparative advantage in moneylending. For our purposes the cause of this shift is irrelevant.


of the Black Death. It thus made sense for the rulers of the larger and better connected cities toprotect their community as Jews had skills and financial capital useful to their resurgence, hence,the accentuation of the protective effects observed in Table 9 rows 3-4 (-0.013*** and -0.010***).

Trade Networks. The presence of Jews might have been more important for land-based traderather than sea-based trade. Credit was vital for trade in medieval Europe as it was highly costlyto move bullion over long distances. Merchants relied on credit to fund trade (see Spufford, 1988).From the 12th century, merchants began to innovate contractual forms to evade the prohibitionon usury, many of which facilitated borrowing money for seaborne trade (Koyama, 2010). Thesea-loan, or foenun nauticum, allowed lenders to earn a return about the principle to compensatefor the risks involved in sea voyage (Hoover, 1926). Thus, the relative importance of the financialintermediation provided by the Jews would have been lower in cities engaged in long-distanceseaborne trade. In these cities, merchants had sophisticated financial contracts that would enablethem to lend and borrow without reliance on Jewish traders. This was not yet the case for inlandtrade as local merchants relied on Jewish moneylenders for credit.

Consistent with this, we find no differential effects for towns along the coast or rivers (rows 5-6).But the protective effect of higher mortality is accentuated for towns at the intersection of twomedieval routes (row 7, -0.009*). We also find a negative effect for towns at the intersection of twomajor Roman roads (row 8, -0.007, but not significant). We find evidence of a stronger protectiveeffect in Jewish communities that were better connected to other communities, using our Jewishcentrality index (row 9, -0.010**). We do not find the same protective effect for towns that werepart of the Hanseatic League—a league of cities that dominated trade in the Baltic region (row 10).This is unsurprising since Jews were not involved in this trading network.

5.4.3. Economic Consequences of Persecutions

To provide further evidence for the complementarities effect, we study simple correlationsbetween population growth and the presence of Jews in, or their disappearance from, the town.

Jewish Presence and Growth. We have data on Jewish presence and population for 1,869 towns⇥ 9 periods from 1100-1850, hence 16,821 observations.60 For town i in year t, we regress logpopulation (log L) in t on a dummy equal to one if Jews are present at any point (J) in [t-1; t]:

logLi,t = Ji,[t�1;t] + �tMi,1347�52 + ⇠i + �t + µi,t . (5)

We include town fixed effects (⇠i) and year fixed effects (�t), so the identifying variation comesfrom towns for which the Jewish presence dummy J changes, either because Jews enter the townor leave the town following a persecution (Jews rarely leave a town on their own). thus measuresthe effect of changes in Jewish presence. Standard errors are clustered at the town level.

60Periods: 1100-1200, 1200-1300, 1300-1400, 1400-1500, 1500-1600, 1600-1700, 1700-1750, 1750-1800 and 1800-1850.


Note that we control for the Black Death mortality rate (M ) interacted with year fixed effects.Indeed, we show that mortality affected persecution probability, whereas Jebwab et al. (2016)show that mortality had long-term effects on city growth, hence the need to control for it. Also,since mortality is only a control variable here, we use the extrapolated Black Death mortality ratesto avoid losing observations (see Section 4.3. for details on how we obtain these). Lastly, theseeffects should be interpreted with caution as Jewish presence is likely endogenous. Also note thatmortality is not a valid instrument for persecutions since it may also affect future city growth.

In row 1 of Table 10, Panel A, we show that towns with a Jewish community were growing 33%faster (over a period of about one century). The point estimates decrease, but remain high andsignificant at 24%, when we add the lag of log population in t-1 (row 2) in order to control formean reversion.61 In row 3, we show that the effect of the Jewish presence dummy is reducedwhen we add the share of years in which Jews were present in [t-1; t] and whose effect is large(0.26***). In row 4, we add a dummy for whether Jews were present at any point in the previousperiod [t-2;t-1] and interact it with our main variable of interest, the Jewish presence dummy in[t-1;t-1]. In doing so, we estimate separate effects for entries and exits (primarily because of apersecution). The effect of Jewish presence is stronger for entries (0.35***, row 4) than for exits(0.12***, so -0,12*** if Jews leave, significantly different at 1%). In row 5, we add a dummy forwhether the year t is strictly after the 14th century, so t > 1400 (and [t-1;t] > [1300;1400]) andinteract it with our main variable of interest. In this way we can estimate separate effects for theeras before/during and after the Black Death. We find that the effect of Jewish presence was notsignificantly different between the two eras (row 5, 0.31*** vs. 0.40***).62

Black Death Persecutions and Growth. For the same panel sample of 16,821 observations, andfor town i in year t, we regress log population (log L) in year t on a dummy Pi,1347�52 equal toone if there has been a persecution in the town during the Black Death period (1347-1352). Sincethis dummy is time-invariant, we interact it with a dummy (t > 1400) for whether the year t isstrictly after the 14th century, so t > 1400 (and thus [t-1;t] > [1300;1400]):

logLi,t = Pi,1347�52 ⇥ (t > 1400) + ⇡Ji,1347�52 ⇥ (t > 1400)

+

0Ji,[t�1;t] + �

0tMi;1347�52 + ⇠

0i + �

0t + µ

0i,t .

(6)

is our main coefficient of interest and measures the effects of persecutions on town growth inthe centuries following the Black Death. Since we use the full sample of 1,869 towns, includingtowns without a Jewish community circa 1347, we control for a dummy Ji,1347�52 equal to one ifJews were present in the town during the Black Death period, also interacted with the post-1400dummy. Similar to equation 5, we add town fixed effects (⇠i), year fixed effects (�t), the Jewish

61In a separate paper, Johnson and Koyama (2017) use several identification strategies to further investigate the roleof Jews in European city growth in 1400-1850, and find similar growth effects of 30% per century.

62Web Appendix Table A.11 shows baseline results hold if we use: (i) two lags of log population, to control for pastgrowth trends, (ii) drop towns not in the Bairoch sample; and (iii) do not replace by 500 inhabitants the population ofthe towns with a missing population, presumably because they were below the 1,000 threshold used by Bairoch.


presence dummy (J) in [t-1; t], and extrapolated mortality (M ) interacted with year fixed effects.

Row 1 of Panel B shows the effect of the Black Death persecution dummy after 1400 (relative tobefore 1400). The effect is negative, at -0.21*, implying that towns where Jews were persecuted in1347-1352 grew relatively slower in the following centuries. An effect size of -21% is close to theeffect of +33% for Jewish presence (see row 1 of Panel A). In row 2 of Panel B, we interact dummiesequal to one if there had been a pogrom or an expulsion during the Black Death period with thepost-1400 dummy (using the same framework as equation 6). The negative correlation betweenpersecution and future growth is especially apparent when the persecution is a pogrom ratheran expulsion (-0.31*** vs. 0.24, significantly different at 1%). By expelling instead of annihilatingtheir community, towns may have been able to re-attract Jewish settlement.63 Consistent with thecomplementarities effect, towns with Jewish communities appear to have grown faster and townswhich persecuted their communities appear to have grown slower in subsequent centuries.

6. Concluding Discussion

We use the Black Death to study how a major shock affected the incentive to persecute Jewishcommunities in Europe. In theory, shocks can increase both the incentive to persecute a minorityand simultaneously raise their economic value. Ultimately, the decision to persecute the minoritydepends on how the magnitude of the shock interacts with the utility one derives from persecutionand the economic benefits associated with the presence of the minority. In line with thescapegoating hypothesis, we find that the Black Death led to the largest massacres of Jews prior tothe Holocaust. However, at the micro-level, the intensity of the shock had a strong countervailingeffect. Cities that were hit especially hard faced a demographic and economic crisis and were lesslikely to persecute their communities. We provide evidence that this micro-level variation can bestbe explained in terms of (i) a town’s cultural inheritance and ideological predilection to scapegoatJews, especially in a context where conspiracy theories abounded, and (ii) the economic role Jewsplayed in a town, although we cannot rule out other explanations. These results suggest that whenthere are latent biases against minorities, shocks can lead to these biases manifesting themselves aspersecutions, with conspiracy theories one of mechanisms through which scapegoating spreads.However, when the minority and majority communities engage in economically complementaryactivities, then these relationships may be a powerful way to reduce inter-group conflict.

REFERENCESAberth, John, From the Brink of the Apocalypse: Confronting Famine, War, Plague, and Death in the Later Middle Ages, London:

Routledge, 2009.Acemoglu, Daron and James A. Robinson, Why Nations Fail, New York: Crown Business, 2012.

, Tarek A. Hassan, and James A. Robinson, “Social Structure and Development: A Legacy of the Holocaust in Russia,”The Quarterly Journal of Economics, 2011, 126 (2), 895–946.

Adena, Maja, Ruben Enikolopov, Maria Petrova, Veronica Santarosa, and Ekaterina Zhuravskaya, “Radio and the Rise ofThe Nazis in Prewar Germany,” The Quarterly Journal of Economics, 2015, 130 (4), 1885–1939.

63In Web Appendix Table A.12, we interact the persecution dummy, or the pogrom and expulsion dummies, withyear fixed effects to measure their effects in each period. Results are then also unchanged.


Adler, Cyrus and Isidore Singer, eds, The Jewish Encyclopedia Funk and Wagnalls New York 1906.Allcott, Hunt and Matthew Gentzkow, “Social Media and Fake News in the 2016 Election,” Journal of Economic Perspectives,

Spring 2017, 31 (2), 211–236.Almond, Douglas, “Is the 1918 Influenza Pandemic Over? Long-Term Effects of In Utero Influenza Exposure in the Post-

1940 U.S. Population,” Journal of Political Economy, 2006, 114 (4), 672–712.Anderson, R. Warren, Noel D. Johnson, and Mark Koyama, “Jewish Persecutions and Weather Shocks 1100–1800,” Economic

Journal, 2016, Forthcoming.Arbath, Cemal Eren, Quamrul H. Ashraf, and Oded Galor, “The Nature of Conflict,” CESifo Working Paper Series 5486,

CESifo Group Munich 2015.Arbatli, Cemal Eren, Quamrul H. Ashraf, and Oded Galor, “The Nature of Conflict,” NBER Working Papers 21079, National

Bureau of Economic Research, Inc April 2015.Archer, John, “The Structure of Anti-Semitism in the ”Prioress’s Tale”,” The Chaucer Review, 1984, 19 (1), 46–54.Backman, Clifford R., The Decline and Fall of Medieval Sicily: Politics, religion, and economy in the reign of Frederick III, 1296-1337,

Cambridge: Cambridge University Press, 1995.Bairoch, Paul, Cites and Economic Development, from the dawn of history to the present, Chicago: University of Chicago Press,

1988. translated by Christopher Braider.Baron, Salo, A Social and Religious History of the Jews, Vol. XI: Citizen or Alien Conjurer, New York: Columbia University

Press, 1967.BBC, There Are Diseases Hidden in Ice, and They Are Waking Up, London, UK: BBC of May 4, 2017.Becker, Sascha O. and Luigi Pascali, “Religion, Division of Labor and Conflict: Anti-Semiticsm in German Regions over 600

Years,” Technical Report 288, Centre for Competitive Advantage in the Global Economy April 2016.Behar, Doron M., Bayazit Yunusbayev, Mait Metspalu, Ene Metspalu, Saharon Rosset, Ju Parik, Siiri Rootsi, Gyaneshwer

Chaubey, Ildus Kutuev, Guennady Yudkovsky, Elza K. Khusnutdinova, Oleg Balanovsky, Ornella Semino, Luisa Pereira,David Comas, David Gurwitz, Batsheva Bonne-Tamir, Tudor Parfitt, Michael F. Hammer, Karl Skorecki, and RichardVillems, “The genome-wide structure of the Jewish people,” Nature, Jul 08 2010, 466 (7303), 238–42.

Ben-Sasson, Haim Hillel, Yehuda Slutsky, and Dina Porat, “Blood Libel,” in Michael Berenbaum and Fred Skolnik, eds.,Encyclopedia Judaica, Vol. 17 Macmillan Reference New York 2007, pp. 774–780.

Benedictow, Ole J., The Black Death 1346–1353: The Complete History, Woodbridge: The Boydell Press, 2005., What Disease was Plague? On the Controversy over the Microbiological Identity of Plague Epidemics of the Past, Leiden: Brill,2010.

Berenbaum, Michael and Fred Skolnik, Encyclopedia Judaica, 2nd ed., USA: Macmillan Reference, 2007.and , “Nuremberg,” in Michael Berenbaum and Fred Skolnik, eds., Encyclopaedia Judaica, Vol. 15 Macmillan Reference2007, pp. 346–348.

Berngruber, Thomas W., Rmy Froissart, Marc Choisy, and Sylvain Gandon, “Evolution of Virulence in EmergingEpidemics,” PLoS Pathog, 03 2013, 9 (3), 1–8.

Biraben, Jean-Noel, La Peste dans l’Histoire: Les Hommes et la Peste en France et dans les Pays Europeens et Mediterraneens, Vol. I,Mouton, Paris, 1975.

Blumenkranz, Bernhard, David Weinberg, and Georges Levitte, “Toulouse,” in Michael Berenbaum and Fred Skolnik, eds.,Encyclopaedia Judaica, Vol. 20 Macmillan Reference New York 2007, pp. 73–74.

Boccaccio, Giovanni, The Decameron, London: Dodo Press, 2005, 1371. Translated by J.M. Rigg.Boerner, Lars and Battista Severgnini, “Epidemic Trade,” 2014. LSE Economic History Working Paper No. 212/2014.Bonnier, Evelina, Jonas Poulsen, Thorsten Rogall, and Miri Stryjan, “Preparing for Genocide: Community Work in

Rwanda,” Working Paper Series 2015:1, Uppsala University, Department of Economics January 2015.Bosker, Maarten, Eltjo Buringh, and Jan Luiten van Zanden, “From Baghdad to London: unravelling urban development in

Europe and the Arab world 800-1800,” Review of Economics and Statistics, 2013, 95 (4), 1418–1437.Botticini, Maristella, “A Tale of ’Benevolent’ Governments: Private Credit Markets, Public Finance, and the Role of Jewish

Lenders in Medieval and Renaissance Italy,” Journal of Economic History, 2000, 60, 164–189.and Zvi Eckstein, The Chosen Few, Princeton, New Jersey: Princeton University Press, 2012.

Campbell, Bruce M.S., The Great Transition: Climate, Disease and Society in the Late-Medieval World, Cambridge: CambridgeUniversity Press, 2016.

Caselli, Francesco and Wilbur John Coleman, “On The Theory Of Ethnic Conflict,” Journal of the European EconomicAssociation, 01 2013, 11, 161–192., Massimo Morelli, and Dominic Rohner, “The Geography of Interstate Resource Wars,” The Quarterly Journal of Economics,2015, 130 (1), 267–315.

Chandler, Tertius, Four Thousand Years of Urban Growth, Lewiston, NY: St David University Press, 1987.Chanes, Jerome A., Antisemitism, Denver, Colorado: Contemporary World Issues, 2004.Chazan, Robert, European Jewry and the First Crusade, Berkeley: University of California Press, 1987.

, Reassessing Jewish Life in Medieval Europe, Cambridge: Cambridge University Press, 2010.Christakos, George, Richardo A. Olea, Marc L. Serre, Hwa-Lung Yu, and Lin-Lin Wang, Interdisciplinary Public Health

Reasoning and Epidemic Modelling: The Case of Black Death, Berlin: Springer, 2005.Chua, Amy, World on Fire: How Exporting Free Market Democracy Breeds Ethnic Hatred and Global Instability, New York: Anchor

Books, 2004.


Cohen, Jeramy, The Friars and the Jews, Ithaca: Colombia University Press, 1982.Cohen, Norman, The Pursuit of the Millenium, revolutionary millenarians and mystical anarchists of the middle ages, London:

Pimilco, 1957.Cohn, Samuel K., “The Black Death and the Burning of Jews,” Past & Present, 2007, 196 (1), 3–36.

and Guido Alfani, “Households and Plague in Early Modern Italy,” The Journal of Interdisciplinary History, 2007, 38 (2),pp. 177–205.

Cremieux, A., “Les Juifs de Toulon au Moyen Age et le massacre du 14 Avril 1348,” Journal of Jewish Studies, 33–72 1930, 89.D’Acunto, Francesco, Marcel Prokopczuk, and Michael Weber, “Historical Antisemitism, Ethnic Specialization, and

Financial Development,” August 2017. memo.de Roover, Raymond, The Rise and Decline of the Medici Bank, 1397–1494, Cambridge, MA: Harvard University Press, 1963.DellaVigna, Stefano and Ethan Kaplan, “The Fox News Effect: Media Bias and Voting,” The Quarterly Journal of Economics,

2007, 122 (3), 1187–1234.Dollinger, Philippe, The German Hansa, London: MacMillan, 1970. translated and edited by D.S. Ault and S.H. Steinberg.Donaldson, David, “Forthcoming.“Railroads of the Raj: Estimating the impact of transportation infrastructure.”,” American

Economic Review, 2018.Donaldson, Terence L., Jews and Anti-Judaism in the New Testament, Causton Street: Baylor University Press, 2010.Duflo, Esther, Pascaline Dupas, and Michael Kremer, “Education, HIV, and Early Fertility: Experimental Evidence from

Kenya,” American Economic Review, September 2015, 105 (9), 2757–2797.Easterly, William, Roberta Gatti, and Sergio Kurlat, “Development, democracy, and mass killings,” Journal of Economic

Growth, 2006, 11 (2), 129–156.Eidelberg, Shlomo, The Jews and the Crusades, Madison, Wisconsin: The University of Wisconsin Press, 1977.Enikolopov, Ruben, Maria Petrova, and Ekaterina Zhuravskaya, “Media and Political Persuasion: Evidence from Russia,”

American Economic Review, December 2011, 101 (7), 3253–85.Epstein, Stephan R., An Island for itself: Economic development and social change in late medieval Sicily, Cambridge: Cambridge

University Press, 1992.Epstein, Steven, Genoa and the Genoese, 958-1528, Chapel Hill: University of North Carolina Press, 1996.Esteban, Joan and Debraj Ray, “On the Salience of Ethnic Conflict,” American Economic Review, December 2008, 98 (5),

2185–2202.and , “Linking Conflict to Inequality and Polarization,” American Economic Review, June 2011, 101 (4), 1345–1374.

, Laura Mayoral, and Debraj Ray, “Ethnicity and Conflict: An Empirical Study,” American Economic Review, June 2012,102 (4), 1310–1342., Massimo Morelli, and Dominic Rohner, “Strategic Mass Killings,” Journal of Political Economy, 2015, 123 (5), 1087–1132.

Field, Erica, Matthew Levinson, Rohini Pande, and Sujata Visaria, “Segregation, Rent Control, and Riots: The Economics ofReligious Conflict in an Indian City,” American Economic Review, May 2008, 98 (2), 505–510.

Freudmann, Lillian C., AntiSemitism in the New Testament, Lanham: University Press of America, 1994.Friedlander, Saul, Nazi Germany and the Jews, Vol. I, New York: Harper Collins, 1997.Galor, Oded, “From Stagnation to Growth: Unified Growth Theory,” in Philippe Aghion and Steven Durlauf, eds., Handbook

of Economic Growth, Elsevier, 2005., Unified Growth Theory, Princeton, New Jersey: Princeton University Press, 2011.

Gensini, Gian Franco, Magdi H. Yacoub, and Andrea A. Conti, “The concept of quarantine in history: from plague to SARS,”Journal of Infection, 2004, 49 (4), 257 – 261.

Gentzkow, Matthew and Jesse M. Shapiro, “Media Bias and Reputation,” Journal of Political Economy, April 2006, 114 (2),280–316.

Girvan, Michelle, Duncan S. Callaway, M. E. J. Newman, and Steven H. Strogatz, “Simple model of epidemics with pathogenmutation,” Phys. Rev. E, Mar 2002, 65, 031915.

Glick, Peter, “Sacrificial Lambs Dressed in Wolves’ Clothing,” in Leonard S. Newman and Ralf Erber, eds., UnderstandingGenocide,, Oxford University Press Oxford 2002., “Choice of Scapegoats,” in John F. Dovidio, Peter Glick, and Laurie A. Rudman, eds., On the Nature of Prejudice: 50 Yearsafter Allport,, Blackwell Malden, MA 2005, pp. 244–61.

Goldhagen, Daniel, The Devil That Never Dies, New York: Little Brown and Company, 2013.Gottfried, Robert S., The Black Death: Natural and Human Disaster in Medieval Europe, New York: The Free Press, 1983.Gottheil, Richard and S. Kahn, “Grenoble,” in Isidore Singer, ed., Jewish Encyclopedia, 1906, pp. 91–92.Grosfeld, Irena, Alexander Rodnyansky, and Ekaterina Zhuravskaya, “Persistent Antimarket Culture: A Legacy of the Pale

of Settlement after the Holocaust,” American Economic Journal: Economic Policy, 2013, 5 (3), 189–226., Seyhun Orcan Sakalli, and Ekaterina Zhuravskaya, “Middleman Minorities and Ethnic Violence: Anti-Jewish Pogromsin the Russian Empire,” 2017. Mimeo.

Haddock, David and Lynne Kiesling, “The Black Death and Property Rights,” The Journal of Legal Studies, 2002, 31 (S2),545–587.

Herlihy, David, “Plague and Social Change in Rural Pistoia, 1201-1430,” The Economic History Review, 1965, 18 (2), 225–244.Hoover, Calvin B., “The Sea Loan in Genoa in the Twelfth Century,” The Quarterly Journal of Economics, 1926, 40 (3), 495–529.Horrox, Rosmery, ed., The Black Death Manchester University Press Manchester 1994.


Jebwab, Remi, Noel D. Johnson, and Mark Koyama, “Bones, Bacteria and Break Points: The Heterogeneous Spatial Effectsof the Black Death and Long-Run Growth,” June 2016. Working Paper.

Jha, Saumitra, “Maintaining peace across ethnic lines: New lessons from the past,” Economics of Peace and Security Journal,2007, 2 (2), 89–93., “Trade, complementaries and religious tolerance: evidence from India,” American Journal of Political Science, 2013, 107(4)., “Trading for Peace,” Economic Policy, 2018, Forthcoming.

Johnson, Noel D. and Mark Koyama, “Jewish Communities and City Growth in Preindustrial Europe,” Journal ofDevelopment Economics, 2017, Forthcoming.

Koyama, Mark, “Evading the ‘Taint of Usury’: The usury prohibition as a barrier to entry,” Explorations in Economic History,2010, 47 (4), 420–442.

Kunda, Ziva, “The case for motivated reasoning.,” Psychological bulletin, 1990, 108 (3), 480.Mayoral, Laura and Debraj Ray, “Groups in Conflict: Size Matters, But Not In The Way You Think,” 2016. Mimeo.McCormick, Michael, Guoping Huang, Giovanni Zambotti, and Jessica Lavash, “Roman Road Network,” 2013. DARMC

Scholarly Data Series 2013-5. Center for Geographic Analysis, Harvard University, Cambridge MA 02138.Michalopoulos, Stelios and Elias Papaioannou, “The Long-Run Effects of the Scramble for Africa,” American Economic

Review, July 2016, 106 (7), 1802–48.Miguel, Edward, “Poverty and Witch Killing,” Review of Economic Studies, October 2005, 72 (4), 1153–1172.Mollat, G., The Popes at Avignon: The “Babylonian Captivity” of the Medieval Church, New York: Harper Touch Books, 1963.Montalvo, Jose G. and Marta Reynal-Querol, “Ethnic Polarization, Potential Conflict, and Civil Wars,” American Economic

Review, June 2005, 95 (3), 796–816.and , “Discrete Polarisation with an Application to the Determinants of Genocides,” Economic Journal, November

2008, 118 (533), 1835–1865.Moore, Robert, The Formation of a Persecuting Society: power and deviance in Western Europe, 950–1250, Oxford: Blackwell,

1987.Moriceau, Jean-Marc, Chroniques paysannes: du Moyen Age au XXe siecle, France Agricole Editions, 2010.Mueller, Reinhold C., The Venetian Money Market, banks, panics, and the public debt, 1200–1500, Baltimore, Maryland: The

Johns Hopkins University Press, 1997.Munro, John H., “Before and after the Black Death: money, prices, and wages in fourteenth-century England,” MPRA Paper

15748, University Library of Munich, Germany December 2004.New York Times, Violence Against Homosexuals Rising, Groups Seeking Wider Protection Say, New York, U.S.A: New York

Times of Nov. 23, 1986., The Ebola Conspiracy Theories, New York, U.S.A: New York Times of Oct. 19, 2014, 2014., Mahmoud Abbas Claims Rabbis Urged Israel to Poison Palestinians’ Water, New York, U.S.A: New York Times of June 23,2016.

Newsweek, #PizzaGate Resurfaces an Old Anti-Semitic Slander, New York, U.S.A: Newsweek of Dec. 6, 2016, 2016.Nohl, Johannes, The Black Death: A Chronicle of the Plague, George Allen & Unwin LTD, 1924. translated by C.H. Clarke.Noonan, John T., The Scholastic Analysis of Usury, Cambridge, Massachusetts, U.S.A: Harvard University Press, 1957.Oster, Emily, “Witchcraft, Weather and Economic Growth in Renaissance Europe,” Journal of Economic Perspectives, Winter

2004, 18 (1), 215–228., “Sexually Transmitted Infections, Sexual Behavior, and the HIV/AIDS Epidemic,” The Quarterly Journal of Economics,2005, 120 (2), 467–515.

Pamuk, Sevket, “The Black Death and the origins of the ‘Great Divergence’ across Europe, 1300-1600,” European Review ofEconomic History, 2007, 11 (3), 289–317.

Pascali, Luigi, “Banks and Development: Jewish Communities in the Italian Renaissance and Current EconomicPerformance,” Review of Economics and Statistics, 2016, 98 (1), 140–158.

Ramirez, Olivier, Les Juifs et le credit en Savoie au XIVe siecle, Vol. 38, Centre Europeen d’Etudes Bourguignonnes, 2010.Ray, Debraj and Joan Esteban, “Conflict and Development,” 2016. Mimeo.Renouard, Yves, The Avignon Papacy, Hamden, Connecticut: Archon Books, 1970. translated by Denis Bethell.Rogall, Thorsten, “Mobilizing the Masses for Genocide,” 2015. Mimeo.

and David Yanagizawa-Drott, “The Legacy of Political Mass Killings: Evidence from the Rwandan Genocide,” 2016.Mimeo.

Rohner, Dominic, Mathias Thoenig, and Fabrizio Zilibotti, “Seeds of distrust: conflict in Uganda,” Journal of EconomicGrowth, September 2013, 18 (3), 217–252., , and , “War Signals: A Theory of Trade, Trust, and Conflict,” Review of Economic Studies, 2013, 80 (3), 1114–1147.

Roosen, J. and D.R. Curtis, “Dangers of Noncritical Use of Historical Plague Data,” Emerging Infectious Diseases, 2018,Forthcoming.

Roth, Cecil, “The Economic History of the Jews,” The Economic History Review, 1961, 14, 131–135.Rubin, Miri, “Desecration of the Host: The Birth of an Accusation,” Studies in Church History, 1992, 29, 169–185.

, Gentile Tales: The Narrative Assault on Late Medieval Jews, Philadelphia: University of Pennsylvania Press, 2004.Russell, Josiah Cox, Medieval Regions and their Cities, Newton Abbot, Devon: David & Charles, 1972.


Satyanath, Shanker, Nico Voigtlander, and Hans-Joachim Voth, “Bowling for fascism: Social capital and the rise of the NaziParty in Weimar Germany, 1919-33,” 2017. Forthcoming in the Journal of Political Economy.

Schmid, Boris V., Ulf Buntgen, W. Ryan Easterday, Christian Ginzler, Lars Walløe, Barbara Bramanti, and Nils Chr. Stenseth,“Climate-driven introduction of the Black Death and successive plague reintroductions into Europe,” Proceedings of theNational Academy of Sciences, 2015, 112 (10), 3020–3025.

Schwarzfuchs, Simon R., “Crusades,” in Michael Berenbaum and Fred Skolnik, eds., Encyclopedia Judaica, Vol. 17 MacmillanReference New York 2007, pp. 310–315.

Seldin, Michael F, Russell Shigeta, Pablo Villoslada, Carlo Selmi, Jaakko Tuomilehto, Gabriel Silva, John W Belmont,Lars Klareskog, and Peter K Gregersen, “European Population Substructure: Clustering of Northern and SouthernPopulations,” PLOS Genetics, 09 2006, 2 (9), 1–13.

Shepherd, William R., Historical Atlas, New York: Henry Holt and Company, 1923.Shirk, Melanie V, “The Black Death in Aragon, 1348–1351,” Journal of Medieval History, 1981, 7 (4), 357–367.Sibon, Juliette, Les juifs de Marseille au XIVe siecle, Paris: CERF, 2011.Spufford, Peter, Money and its use in Medieval Europe, Cambridge: CUP, 1988.Stacey, Robert C., “From Ritual Crucifixion to Host Desecration: Jews and the Body of Christ,” Jewish History, 1998, 12 (1),

11–28.Stasavage, David, “Was Weber Right? The Role of Urban Autonomy in Europe’s Rise,” American Political Science Review, 5

2014, 108, 337–354.Talbert, Richard J. A, ed., Barrington Atlas of the Greek and Roman World Princeton University Press Princeton 2000.Theilmann, John and Frances Cate, “A Plague of Plagues: The Problem of Plague Diagnosis in Medieval England,” The

Journal of Interdisciplinary History, 2007, 37 (3), 371–393.Trachtenberg, Joshua, The Devil and the Jews, Philadelphia: The Jewish Publication Society, 1943.van Straten, Jits, “Early Modern Polish Jewry The Rhineland Hypothesis Revisited,” Historical Methods, 2007, 40 (1), 39–50.Vigna, Stefano Della, Ruben Enikolopov, Vera Mironova, Maria Petrova, and Ekaterina Zhuravskaya, “Cross-Border Media

and Nationalism: Evidence from Serbian Radio in Croatia,” American Economic Journal: Applied Economics, July 2014, 6(3), 103–132.

Voigtlander, Nico and Hans-Joachim Voth, “Persecution Perpetuated: The Medieval Origins of Anti-Semitic violence inNazi Germany,” Quarterly Journal of Economics, 2012, 127 (3), 1–54.and , “Married to Intolerance: Attitudes toward Intermarriage in Germany, 1900-2006,” American Economic Review,

May 2013, 103 (3), 79–85.and , “The Three Horsemen of Riches: Plague, War, and Urbanization in Early Modern Europe,” Review of Economic

Studies, 2013, 80 (2), 774–811.and , “Highway to Hitler,” ECON - Working Papers 156, Department of Economics - University of Zurich May 2014.and , “Nazi indoctrination and anti-Semitic beliefs in Germany,” Proceedings of the National Academy of Sciences, 2015,

112 (26), 7931–7936.Waldinger, Fabian, “Quality Matters: The Expulsion of Professors and the Consequences for PhD Student Outcomes in Nazi

Germany,” Journal of Political Economy, 2010, 118 (4), 787–831., “Peer Effects in Science: Evidence from the Dismissal of Scientists in Nazi Germany,” Review of Economic Studies, 2012,79 (2), 838–861., “Bombs, Brains, and Science: The Role of Human and Physical Capital for the Creation of Scientific Knowledge,” TheReview of Economics and Statistics, December 2016, 98 (5), 811–831.

Wasserman, Henry, “Regensburg,” in Michael Berenbaum and Fred Skolnik, eds., Encyclopedia Judaica, Vol. 17 MacmillanReference New York 2007, pp. 188–189.

Yanagizawa-Drott, David, “Propaganda and Conflict: Evidence from the Rwandan Genocide,” The Quarterly Journal ofEconomics, 2014, 129 (4), 1947–1994.

Young, Alwyn, “The Gift of the Dying: The Tragedy of Aids and the Welfare of Future African Generations,” The QuarterlyJournal of Economics, 2005, 120 (2), 423–466.

Zanden, Jan Luiten Van, Eltjo Buringh, and Maarten Bosker, “The Rise and Decline of European Parliaments, 1188–1789,”Economic History Review, 08 2012, 65 (3), 835–861.

Ziegler, Philip, The Black Death, London: Collins, 1969.


Figure 1: Cumulative Black Death Mortality Rates (%) in 1347-1352

Mortality Cities (263)0-20

20-40

40-60

60-80

80-100

Messina

Genoa

To Kaffa

Notes: This map depicts Black Death mortality rates (%, 1347-1352) for 263 localities in 13 Western European countries:Austria, Belgium, the Czech Republic, France, Germany, Ireland, Italy, Norway, Portugal, Spain, Sweden, Switzerlandand the United Kingdom. The main source for the mortality data is Christakos et al. (2005).

Figure 2: Main Sample of 124 Towns with a Jewish Community and Mortality Data

DD

DD

D

D

D

DD

D

D

D

D

D

D

D

D

D

D

D

DD

DD

D

D

D

D

D

D

DD

DDD D

D

D

D

D

D

D

D

D

D

D

D

D

D

D

D

D

D

D

D

D

D

D

D

D

DD

D

D

D D

D

D

D

D

D D

D

D

DD

D

D

D

D

D

DD

D

D

DD

D

D

D

D

DDDD

DD

D

D

D

DD

D

D

DD

D

D

D

D

D

D

D

D

D

D

D

D

D

DD

DDD

D Jewish Towns (124)

Notes: This map shows the 124 towns of our main sample, i.e. the towns with Jews present circa 1347 and for which weknow the cumulative Black Death mortality rate (%, 1347-1352). These 124 towns belong to 9 countries using today’sboundaries: Austria, Belgium, the Czech Republic, France, Germany, Italy, Portugal, Spain and Switzerland.


Figure 3: Distribution of the Duration of the Black Death, and Relationship between CumulativeBlack Death Mortality and the Likelihood of a Persecution in 1347-1352.

(a) Duration (b) Persecution

Notes: Panel (a) shows the Kernel distribution of the duration of the Black Death in each city—i.e. the time differencebetween the month of the first infection in the city and the month of the last infection in the city (information availablefor 39 out of the 124 towns of the main sample). The mean and median durations were 5 months. Panel (b) showsthat for the main sample of 124 Jewish towns for which we have data on Black Death mortality there is a positive andthen negative correlation between the likelihood of a persecution and the mortality rate in 1347-1352 (local polynomialsmooth plot with a bandwidth of 5 percentage points of mortality). See Web Appendix for more details on data sources.

Figure 4: Cumulative Black Death Mortality, Locality Size, and Market Access in 1300

(a) Locality Size (b) Market Access

Notes: Panel (a) depicts the relationship between mortality rates (%, 1347-1352) and log population size in 1300 for ourmain sample of 124 Jewish towns (Y = 40.36*** -0.55 X; Obs. = 124; R2 = 0.00). Panel (b) depicts the relationship betweenmortality rates (%, 1347-1352) and log market access to all 1,869 towns in 1300 for our main sample of 124 Jewish towns(Y = 38.45*** +0.85 X; Obs. = 124; R2 = 0.01). Market access for city i is defined as MAi = ⌃j

Pj

D�ij

, with Pj being thepopulation of town j 6= i, Dij the travel time between city i and city j, and � = 3.8. To obtain the travel times, wecompute the least cost travel paths via four transportation modes — by sea, by river, by road and by walk — with thetransportation speeds from Boerner & Severgnini (2014). See Web Appendix for more details on data sources.


Figure 5: Total Number of Jewish Persecutions in 1100-1600 and 1347-1352

(a) All Persecutions, 1100-1600 (b) Black Death Persecutions, 1347-1352

Notes: Panel (a) depicts all persecutions recorded in the full sample of 1,869 towns when the year is rounded to thenearest decade (year ending in 0) or mid-decade (year ending in 5). It shows that the Black Death period (1347-1352,rounded in 1350) witnessed the greatest number of persecutions in medieval European history (here, 1100-1600). Panel(b) focuses on the 124 towns of the main sample and on persecutions that took place within the Black Death period andseven years before or after (here, 1340-1360). See Web Appendix for more details on data sources.

Figure 6: Market Access to All Towns vs. Market Access to Messina Only, 1300.

(a) All Towns (b) Messina Only

Notes: Panel (a) shows for the 124 towns of the main sample their log market access to all towns in 1300. Panel (b)shows for the same 124 towns their log market access to Messina in 1300. See notes under figure 4(b) for details on howmarket access is calculated. We use as an instrument log market access to Messina, conditional on log market access toall 1,869 towns in Western Europe. See Web Appendix for more details on data sources.


Figure 7: Timing of the Onset of the Black Death and Black Death Mortality

Notes: This figure shows for the sample of 124 towns the relationship between cumulative Black Death mortality rates(%) in 1347-1352 and the timing of the onset of the Black Death in the town. Number of months is measured sinceOctober 1347, the date Messina—the port of entry of the Black Death in Europe—was infected. Towns infected earlierhad higher mortality rates (Y = 52.01*** -0.87*** X; R2 = 0.22). See Web Appendix for more details on data sources.

Figure 8: Mapping of First Crusade, Blood and Well-Poisoning Libels, and Financial Competition.

(a) First Crusade, Blood and Well-Poisoning Libels

((

((

(

((( ((

(

(

"

"

""

"

"

"

"""

"

^

^ Chillon Castle

" Host Desecration 13C-14C

Ritual Murder 13C-14C

( 1st Crusade Pogroms Rhine River

(b) Financial Centers and Moneylending

"

" ""

"

"

""

"

"

""

""""

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

"

""

"

"

"

"

"

"

"

"

"

"

"

"

""

"

"

"

"

"

"

"

"

"

"

!

!

!

!

!

!

!

!

!

!!

!

!

!

!!

!

!

! !

!

!!

!

!

!!

!

!

!

!!

!

!!

!!

!

!

!

!

!!

!

!!!!

!

!

!

!

!

!

!

!

!

!!!

##

##

#

#

# Financial Center ! Money Lending " No Money Lending

Notes: Panel (a) shows the locations of the 12 pogroms that took place during the First Crusade (1096), the locations ofthe 19 charges of ritual murder and the 11 charges of host desecration in the 13th and 14th centuries, Chillon (the placeof origin of the well-poisoning libel), and the Rhine river (along which the rumor of a Jewish conspiracy spread). Panel(b) shows the main financial centers of Western Europe (Cahors, Florence, Genoa, Milan, Sienna and Venice) and the 60towns where Jews were lending money ca. 1347. See Web Appendix for more details on data sources.


Table 1: BLACK DEATH MORTALITY RATES AND JEWISH PERSECUTIONS, 1100-1600

Dependent Variable: Dummy if Any Jewish Persecution in Period [t-1; t]:

Mortality in 1347-1352 Constant Obs. R2

1. [t-1; t] = [1347-1352] -0.009*** [0.002] 0.831*** [0.104] 124 0.12

2. [t-1; t] = [1353-1400] -0.004* [0.002] 0.404*** [0.098] 122 0.023. [t-1; t] = [1353-1500] -0.000 [0.002] 0.640*** [0.099] 124 0.004. [t-1; t] = [1353-1600] 0.000 [0.002] 0.724*** [0.088] 127 0.00

5. [t-1; t] = [1341-1346] 0.001 [0.000] -0.004 [0.004] 122 0.016. [t-1; t] = [1321-1346] -0.001 [0.001] 0.144** [0.068] 126 0.017. [t-1; t] = [1300-1346] -0.001 [0.002] 0.255*** [0.082] 131 0.008. [t-1; t] = [1200-1346] -0.003 [0.002] 0.370*** [0.090] 132 0.02

Notes: This table shows the constant ↵t and the effect �t of the Black Death mortality rate (%) in 1347-1352 on a dummy equal to one ifthere has been any persecution in various periods [t-1; t], for the towns for which we have mortality data and in which we know thatJews were present in period [t-1; t], without controls. Robust SE’s: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.

Table 2: TOWN CHARACTERISTICS AND BLACK DEATH MORTALITY RATES

Dependent Variable: Black Death Mortality Rate (%, 1347-1352)

(1) (2) (3) (4)

Average Temperature 1500-1600 (d) -0.23 [0.68] 0.26 [0.83]Elevation (m) -0.01 [0.01] -0.00 [0.01]Cereal Suitability Index -2.28* [1.28] -2.52 [1.52]Pastoral Suitability Index 4.58 [5.25] -0.87 [6.49]Coast 10 Km Dummy -6.67 [4.98] -8.94 [5.82]Rivers 10 Km Dummy -3.15 [3.08] -4.26 [3.85]Latitude (d) -2.39*** [0.58] -1.90** [0.79]Longitude (d) 0.62** [0.30] 1.09** [0.42]

Log Town Population in 1300 -1.41 [1.23] -1.56 [1.57]Maj.Roman Rd (MRR) 10 Km Dummy 0.49 [8.48] -4.74 [6.30]MRR Intersection 10 Km Dummy 9.67* [5.71] 8.62 [5.65]Any Roman Rd (ARR) 10 Km Dummy 6.78 [9.23] 10.39 [7.57]ARR Intersection 10 Km Dummy -5.54 [5.53] -2.15 [5.49]Medieval Route (MR) 10 Km Dummy 1.07 [4.42] -1.73 [4.06]MR Intersection 10 Km Dummy -3.04 [5.14] -3.94 [5.41]Market and Fair Dummy -5.14 [4.38] -0.97 [5.16]Hanseatic League Dummy -1.03 [6.30] 7.20 [6.88]Log Market Access in 1300 0.51 [1.00] -0.07 [1.06]Aqueduct 10 Km Dummy 3.19 [4.22] -0.33 [4.66]University Dummy 3.88 [6.47] 4.43 [7.02]

Monarchy in 1300 Dummy 3.18 [5.44] 6.85 [5.51]State Capital in 1300 Dummy 7.06 [6.43] 2.01 [7.45]Parliamentary Activity in 1300-1400 5.08 [4.66] -0.32 [4.59]Log Distance to Nearest Parliament 4.49** [1.95] 0.59 [2.09]Self-Governing City in 1300 Dummy -5.30 [4.04] 2.04 [4.38]Battle w/i 100 Km in 1300-1350 Dummy -3.59 [3.85] -6.48 [4.31]

Obs.; R2 124; 0.27 124; 0.12 124; 0.15 124; 0.36

Notes: Each column is a separate regression. The variables in (1), (2) and (3) proxy for physical geography, trade and humancapital, and institutions, respectively. Robust SE’s: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.


Table 3: MORTALITY RATES AND PERSECUTIONS, INVESTIGATION OF CAUSALITY

Dependent Variable: Dummy if Any Jewish Persecution in 1347-1352:

Mortality 1347-1352 Constant Obs.

1. Baseline (Row 1 of Table 1) -0.009*** [0.002] 0.831*** [0.104] 124

2. Drop if Jews � 5% of Town Population -0.008*** [0.002] 0.779*** [0.113] 1073. Controls for Jewish Cemetery, Quarter and Synagogue -0.009*** [0.002] 0.808*** [0.117] 1244. Controls for Years of First Entry and Last Reentry -0.009*** [0.002] 0.814*** [0.197] 1245. Control for Jewish Centrality Index -0.009*** [0.002] 0.814*** [0.197] 1246. Keep if Year of Last Entry is After 1290 -0.007* [0.004] 0.82*** [0.179] 367. Drop if Known Number of Victims -0.009*** [0.002] 0.775*** [0.111] 1158. Dummy if Persecution in 1321-1346 -0.009*** [0.002] 0.812*** [0.107] 1249. Dummy if Persecution in 1300-1346 -0.009*** [0.002] 0.816*** [0.108] 12410. Control for Number of Persecutions in 1321-1346 -0.009*** [0.002] 0.812*** [0.107] 12411. Control for Number of Persecutions in 1300-1346 -0.009*** [0.002] 0.836*** [0.106] 12412. Drop if Jewish Presence Inferred from Jewish Persecution -0.008*** [0.002] 0.767*** [0.108] 114

13. Drop Top and Bottom 25% in Mortality -0.015** [0.007] 1.049*** [0.295] 7114. Drop if Natural Baths or Response -0.009*** [0.002] 0.812*** [0.106] 121

15. All Controls of Table 2 -0.005* [0.003] 1.373 [1.344] 12416. All Controls Excl. Latitude, Longitude and Temperature -0.006** [0.002] 1.148*** [0.308] 124

Notes: Robust SE’s: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.

Table 4: MORTALITY RATES AND PERSECUTIONS, IV STRATEGIES




2. IV1: Log MA to Messina, Control for Log MA (F: 31.0; 4.42***) -0.016*** [0.005] 1.134*** [0.184] 1233. IV1 + Latitude, Longitude and their Squares (F: 4.3; 3.46**) -0.023* [0.014] 8.356* [4.606] 1234. IV2: #Months b/w Oct 1347 and 1st Infection (F: 33.3; -0.87***) -0.028*** [0.006] 1.567*** [0.240] 1245. IV2 + Latitude, Longitude and their Squares (F: 7.3; -0.83***) -0.029** [0.015] 6.116 [5.522] 1246. IV1 (Messina) + IV2 (#Months) + Lat., Long. and Sq. (F: 4.4) -0.019* [0.011] 8.652** [4.239] 123

7. Reduced-Form Effect of Log MA to Messina, Ctrl for Log MA -0.071*** [0.002] -0.192 [0.200] 1238. Reduced-Form Effect of #Months b/w Oct 1347 and 1st Inf. 0.024*** [0.004] 0.113 [0.069] 124


9. Reduced-Form Effect of Log MA to Messina, Ctrl for Log MA -0.001 [0.007] 0.061 [0.076] 12110. Reduced-Form Effect of #Months b/w Oct 1347 and 1st Inf. -0.002 [0.002] 0.108** [0.047] 122


11. Reduced-Form Effect of Log MA to Messina, Ctrl for Log MA -0.003 [0.013] 0.103 [0.136] 12112. Reduced-Form Effect of #Months b/w Oct 1347 and 1st Inf. -0.004 [0.004] 0.207*** [0.007] 122

Notes: Rows 2-3 instrument mortality by log market access (MA) to Messina, controlling for log market access (MA) to all 1,869 towns inWestern Europe (IV1). Rows 4-5 instrument mortality by the number of months between October 1347 and the month of first infection inthe town (IV2). See Web Appendix Table A.5 for the full 1st-stage regressions. Rows 9-12: Reduced-form effects of the IVs in 1347-1352,1321-1346 and 1300-1346, respectively. Robust SE’s: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.


Table 5: MORTALITY RATES AND PREVENTIVE PERSECUTIONS




2. Add Dummy if Year-Month of Infection � Sept. 1348 -0.007*** [0.002] 0.601*** [0.138] 124

3. Row 2 + Log Dist. to Chillon + Dummy x Log Dist. -0.007*** [0.002] -0.251 [0.899] 1244. Row 2 + Log Dist. to Letters + Dummy x Log Dist. -0.005** [0.003] -0.972 [0.773] 1245. Row 2 + Log Dist. to Rhine + Dummy x Log Dist. -0.005** [0.003] -0.972 [0.773] 1246. Row 2 + Log Dist. to Flagellants + Dummy x Log Dist. -0.007*** [0.002] 0.499*** [0.133] 124

7. Drop if Likely Preventive Based on Year -0.009*** [0.002] 0.806*** [0.107] 1218. Drop if Possibly Preventive Based on Year -0.009*** [0.002] 0.788*** [0.109] 1199. Drop if Likely Preventive Based on Year-Month -0.008*** [0.002] 0.754*** [0.111] 11510. Drop if Possibly Preventive Based on Year-Month -0.008*** [0.002] 0.763*** [0.112] 114

Notes: Row 2: We add a dummy if the town was infected from Sept. 1348. Rows 3-6: We also control for the log Euclidean distance toChillon, towns that first received letters warning them of the conspiracy, the Rhine, and the path of the flagellants, and their interactionwith the post-Sept. 1348 dummy. Rows 7-10: We drop towns where it is likely/possible that the persecution preceded the Black Death,based on the year/month of first infection. Robust SE’s: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.

Table 6: MORTALITY AND PERSECUTIONS, ALTERNATIVE OUTCOMES

Dependent Variable: Dummy if . . . Mortality 1347-1352 Constant Obs. (Sample)

1. Baseline: Persecution (N = 58) -0.009*** [0.002] 0.831*** [0.104] 124 (All)

2. Pogrom (N = 53) -0.007*** [0.002] 0.720*** [0.105] 124 (All)3. Expulsion (N = 13) -0.004** [0.001] 0.244*** [0.077] 124 (All)

4. Expulsion or Annihilation (N = 32) -0.009** [0.004] 0.837*** [0.137] 58 (Persecution)5. Annihilation (N = 19) -0.008* [0.004] 0.694*** [0.166] 45 (Pogrom Only)6. Burning (N = 8) -0.003 [0.002] 0.178 [0.108] 45 (Pogrom Only)7. Mob Involved (N = 11) -0.010*** [0.003] 0.556*** [0.160] 45 (Pogrom Only)8. Annihilation, Burning or Mob (N = 28) -0.010** [0.004] 0.844*** [0.157] 45 (Pogrom Only)

9. Persecution + Successful Prevention (N = 61) -0.009*** [0.002] 0.857*** [0.103] 124 (All)10. Persecution - Failed Prevention (N = 50) -0.009*** [0.002] 0.751*** [0.103] 124 (All)11. Any Attempt to Prevent (N = 11) 0.003 [0.003] 0.075 [0.086] 61 (Row 9)

12. Drop Top 25% in Mortality (Mortality � 50%) -0.009*** [0.003] 0.835*** [0.128] 101 (All)13. Keep if Population 1300 < Median (11,000) -0.007*** [0.003] 0.844*** [0.142] 61 (All)14. Keep if Population 1300 � Median (11,000) -0.012*** [0.003] 0.872*** [0.158] 63 (All)15. Keep if Residual Population < Median (5,600) -0.007*** [0.003] 0.812*** [0.155] 62 (All)16. Keep if Residual Population � Median (5,600) -0.015*** [0.004] 0.959*** [0.156] 62 (All)

Notes: Row 4: The annihilation dummy is equal to one if all Jews are killed. Row 6: The burning dummy is equal to one if at least oneJew is burned. Row 7: The mob dummy is equal to one if the persecution is initiated by a mob. Rows 9-10: We know if the local authoritytried to prevent a persecution, and succeeded in doing so. Rows 13-16: We restrict the sample to towns for which the population in 1300or the estimated residual population in the immediate aftermath of the Black Death was below or above the median in the sample of124 towns (11,000 and 5,600 inh., respectively). Robust SE’s: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.


Table 7: MORTALITY AND PERSECUTIONS, SAMPLING AND MEASUREMENT



1. Use Town Population in 1300 as Regression Weights -0.011*** [0.003] 0.749*** [0.181] 1242. Use Mortality of Nearest Avail. Town within 100 km -0.007*** [0.000] 0.690*** [0.076] 3073. Mortality Extrapolated from 263 Towns -0.015*** [0.002] 0.963*** [0.075] 3634. Use Provincial Mortality from Christakos et al. (2005) -0.008*** [0.002] 0.722*** [0.090] 159

5. Controlling for Log Town Population 1300 -0.009*** [0.002] 0.921*** [0.116] 1246. Dropping Bottom and Top 25% in Population 1300 -0.008*** [0.003] 0.812*** [0.142] 637. Dropping Top 10% in Population 1300 -0.009*** [0.002] 0.808*** [0.110] 111

8. Use Information from Encyclopedia Judaica Only -0.013*** [0.002] 0.933*** [0.106] 979. Drop if Unsure about Jewish Presence or Persecution -0.012*** [0.002] 0.957*** [0.109] 9210. Use All Towns with Mortality Data -0.007*** [0.001] 0.533*** [0.070] 263

11. Drop Description-Based Mortality Data -0.011*** [0.003] 0.927*** [0.110] 9712. Drop Desertion-Based Mortality Data -0.006** [0.003] 0.624*** [0.131] 9913. Drop Number-Based Mortality Data -0.010*** [0.003] 0.921*** [0.138] 5214. Use Only Number-Based Mortality Data -0.007** [0.003] 0.704*** [0.159] 72

15. Keep Top and Bottom 25% in Mortality -0.008*** [0.002] 0.813*** [0.117] 53

16. Conley standard errors (100 km) -0.009*** [0.002] 0.831*** [0.112] 124


Table 8: MORTALITY AND PERSECUTIONS, EVIDENCE ON SCAPEGOATING


Effect of: Mortality Mortality x SumRows 1-17: Dummy Equal to 1 if: Rate (�) Dummy (�) (� + �)

1. Close to Chillon Castle (Origin of Rumor) -0.010*** [0.002] 0.016* [0.009] 0.006 [0.009]2. Close to Towns Warned by Letter (Path of Rumor) -0.010*** [0.002] 0.017* [0.009] 0.007 [0.009]3. Close to Rhine River (Path of Rumor) -0.009*** [0.002] 0.017** [0.007] 0.008 [0.007]4. Seat of Papacy (Avignon) in 1347 -0.010*** [0.002] 0.018*** [0.006] 0.008 [0.005]5. Seat of Other Bishopric or Archbishopric ca. 1347 -0.006* [0.003] -0.007 [0.005] -0.013*** [0.003]

6. Very Recent Entry ( 5 Years) -0.011*** [0.002] 0.015*** [0.005] 0.005 [0.004]7. Jewish Cemetery, Quarter or Synagogue ca. 1347 -0.009*** [0.003] 0.000 [0.005] -0.009** [0.004]8. Ashkenazi Settlement 13th Century -0.011*** [0.002] 0.015** [0.007] 0.004 [0.007]

9. Very Recent Persecution in 1340-1346 -0.010*** [0.002] 0.039* [0.021] 0.029 [0.021]10. Close to Pogrom 1st Crusade (1096) -0.009*** [0.002] 0.016** [0.007] 0.007 [0.007]11. Close to Alleged Ritual Murder 13C -0.010*** [0.002] 0.016** [0.008] 0.006 [0.007]12. Close to Alleged Host Desecration 1st Half 14C -0.010*** [0.002] 0.011** [0.005] 0.001 [0.005]

13. Month of First Infection is December (Advent) -0.009*** [0.002] -0.022** [0.009] -0.031*** [0.013]14. Month of First Infection is January (Christmastide ) -0.011*** [0.002] 0.016** [0.007] 0.005 [0.007]15. Month of First Infection is February or March (Lent) -0.008*** [0.002] -0.014*** [0.003] -0.022*** [0.002]16. Month of First Infection is April or May (Easter) -0.011*** [0.002] 0.011* [0.006] 0.000 [0.006]17. Month of First Infection is October (2nd Planting) -0.009*** [0.003] 0.011*** [0.004] 0.002 [0.003]

Notes: This table shows the respective effects of the mortality rate (�) and the respective interacted effects (�) of the mortality ratetimes a dummy variable shown in each row on a dummy equal to one if there has been any Jewish persecution in 1347-1352, for themain sample of 124 Jewish towns. It also shows the combination of the effect of the mortality rate (�) and the interacted effect of themortality rate and the dummy (�). Robust SE’s: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.


Table 9: MORTALITY AND PERSECUTIONS, EVIDENCE ON COMPLEMENTARITIES



1. Close to Major Financial Centers -0.008*** [0.002] 0.008*** [0.002] 0.000 [0.000]2. Jews Lend Money in the Town -0.006* [0.003] -0.009** [0.004] -0.014*** [0.002]

3. Top 10% Town Population in 1300 -0.009*** [0.002] -0.013*** [0.004] -0.024** [0.108]4. Top 10% Overall Market Access in 1300 -0.008*** [0.003] -0.010*** [0.004] -0.018*** [0.115]

5. Coast 10 Km -0.009*** [0.003] 0.002 [0.007] -0.010 [0.006]6. Rivers 10 Km -0.008*** [0.003] -0.004 [0.005] -0.011*** [0.004]7. Medieval Route Intersection 10 Km -0.009*** [0.002] -0.009* [0.005] -0.018*** [0.005]8. Major Roman Road Intersection 10 Km -0.004 [0.004] -0.007 [0.005] -0.010*** [0.003]9. Top 10% Jewish Centrality Index -0.008*** [0.002] -0.010** [0.004] -0.018*** [0.004]10. Hanseatic League -0.009*** [0.002] 0.005 [0.005] -0.004 [0.004]

Notes: This table shows the respective effects of the mortality rate (�) and the respective interacted effects (�) of the mortality ratetimes a dummy variable shown in each row on a dummy equal to one if there has been any Jewish persecution in 1347-1352, for themain sample of 124 Jewish towns. It also shows the combination of the effect of the mortality rate (�) and the interacted effect of themortality rate and the dummy (�). Robust SE’s: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.

Table 10: JEWISH PRESENCE, BLACK DEATH PERSECUTIONS, AND TOWN GROWTH

Dependent Variable: Log Town Population in Year t:

Panel A: Effect of Jewish Presence in Period [t-1; t] Dummy Coeff. SE Obs.

1. Baseline Effect of Jewish Presence Dummy 0.33*** [0.04] 16,8212. Row 1 + Including the Lag of Log Town Population in t-1 0.24*** [0.03] 16,821

3.Individual Effect of Jewish Presence Dummy 0.12* [0.06]

16,821Individual Effect of Jewish Presence Share 0.26*** [0.07]

4.Entries: Effect if Jews Absent in Previous Period [t-2; t-1] 0.35*** [0.05]

16,821Exits: Effect if Jews Present in Previous Period [t-2; t-1] 0.12** [0.05]

5.Effect of Jewish Presence Dummy Before 14th Century 0.31*** [0.04]

16,821Effect of Jewish Presence Dummy After 14th Century 0.40*** [0.06]

Panel B: Effect of Black Death Persecutions in 1347-1352 After 14th Century Coeff. SE Obs.

1. Baseline Effect of Persecution in 1347-1352 Dummy After 14th Century -0.21* [0.11] 16,821

2.Effect of Pogrom in 1347-1352 Dummy After 14th Century -0.31*** [0.12]

16,821Effect of Expulsion in 1347-1352 Dummy After 14th Century 0.24 [0.16]

Notes: All regressions include town FE and year FE, and extrapolated Black Death mortality (%) interacted with year FE. Panel A showsthe effect on log town population in year t of a dummy equal to one if Jews were present in the period [t-1;t]. Row 2: We control forlog town population in year t-1. Row 3: We add the share of years in which Jews were present in [t-1;t]. Row 4: We add a dummy forwhether Jews were present in the previous period [t-2;t-1], and interact it with our main variable of interest. Row 5: We add a dummyfor whether the year t if strictly after the 14th century, so t > 1400 (and [t-1;t] > [1300;1400]), and interact it with our main variableof interest. We do not show the effect of each individual dummy. In Panel B, our dependent variable is log town population in year t.We interact a dummy equal to one if there has been a persecution/pogrom/expulsion during the Black Death period (1347-1352) witha dummy for whether the year t if strictly after the 14th century, so t > 1400 (and [t-1;t] > [1300;1400]). We do not show the effect ofeach individual dummy. Robust SE’s clustered at the town level: *** p<0.01, ** p<0.05, * p<0.10. See Web Appendix for data sources.

WEB APPENDIX - NOT FOR PUBLICATION 1

WEB APPENDIX - NOT FOR PUBLICATION

1. Conceptual Framework

We now describe how a shock can affect the probability that an outgroup is persecuted. Percapita income y depends on population L. In line with the historical discussion, it also dependspositively on the value of the economic services provided by the minority x: y = f(L, x).

y = f(L, x) .

The utility of in-group members has three components: income, the death of in-group members �,and preferences over diversity z (i.e. preferences over the out-group).

ui = g

�y, �, z

�

For simplicity, we work with a linear, separable, version of this function (thereby ignoring cross-effects): ui = y+ �+ z. The effect of the Black Death mortality M on the utility of a representativemember of the in-group (who survives the plague) depends on how these three components areaffected.

dui

dM

=dy

dM

+d�

dM

+dz

dM

(1)

The first component, the effect of the Black Death shock on per capita income could be positivedue to the Malthusian effect on per capita income ( dL

dM@y@L ). However, as we describe in the main

text, Black Death mortality disrupted production causing incomes to initially decrease. The secondcomponent ( d�

dM ) is negative: the death of family, friends and others reduces utility. The third effectis context specific. If the minority is held responsible for the shock, then it is negative ( dz

dM ), andthe in-group resents the presence of the out-group more, the larger the shock.

We now consider the incentive of the in-group to persecute or not persecute the minority. Clearly,if the minority group provides no economic services (x = 0), then there will always be an incentiveto persecute the minority so long as they are held responsible for the shock (i.e. dz/dM < 0).However, if the minority provides important economic services whether or not a persecution willtake place depends on the relative size of the effects.

To analyze the decision to persecute, we employ the method of comparative statics. We fix thethreshold level of utility u at which the representative member of the in-group is indifferentbetween persecuting or not persecuting the minority:

u = u

�y(L, x), �, z

�= u

�y(L), �

�(2)

where x and z are set to zero on the RHS as this is their value when there is no minority community.By revealed preference we know that prior to the plague: ui > u for all cities with minoritycommunities. To see how the decision to persecute is affected by the Black Death shock we take

2 WEB APPENDIX - NOT FOR PUBLICATION

the total derivative of this indifference condition with respect to M :

du

dM

=dL

dM

@y(L, x)

@L

+dx

dM

@y(L)

@x

+d�

dM

+dz

dM

� dL

dM

@y

@L

� d�

dM

. (3)

The direct negative effects of plague mortality on utility cancel out (�). Therefore, therepresentative individual will be more likely to persecute the minority if:

��dL

dM

⇣@y(L, x)

@L

� @y(L)

@L

⌘

| {z }Complementarities Effect

�� <��

dz

dM|{z}Scapegoating Effect

�� . (4)

If relative size of the complementarity and scapegoating effects vary from city to city then a givenmortality shock could have differential effects. Moreover for a given set of city characteristics, thesize of the mortality shock might give rise to differential effects. Specifically, it is natural to assumethat the effects of M on z are linear. If the economic role of the minority community becomes moreimportant in the presence of a large mortality shock, then the effects of the mortality shock on u

might be negative for small values of M and become positive for higher values of M . This wouldgenerate an inverted U-shaped relationship between Black Death mortality and the probability ofa persecution.

Example with a Cobb-Douglas Production FunctionFor illustrative purposes, consider a Cobb-Douglas production function: Y = L

↵(1+x)1�↵.Per capita income is therefore: y = L

↵�1(1 + x)1�↵. We also assume a linear relationshipbetween plague mortality and labor: L = �M , and linear expressions for the disutility ofrelatives dying: (� = �⇡M ) and over preferences over diversity (z = � M ).In this case, the indifference condition (equation 2) can be written as:

u = L

↵�1(1 + x)1�↵ � ⇡M � M = L

↵�1 � ⇡M (5)

Totally differentiating this equation with respect to plague mortality M and simplifying, weobtain:

du

dM

= �x

1�↵⇥(↵� 1)L↵�2

⇤� (6)

The first term is the complementarities effect; it is positive. The second is the scapegoatingeffect; it is negative. This is illustrated in Figure A.1.

To summarize, the shock-persecution relationship ultimately depends on the scapegoating andcomplementarity effects. Below, we will use the insights from this framework when exploringeconometrically the mechanisms explaining the observed shock-persecution relationship.


Figure A.1: Relationship between Mortality and the Scapegoating and Complementarities Effects.

@@@@

@@@

@@@

@@@@

MM

⇤

u

�x

1�↵⇥(↵� 1)L↵�2

⇤

�

u

2. The Main Sample of 124 TownsAmong the 1,869 Western European towns that reached 1,000 inhabitants at one point between 800and 1850 in the Bairoch (1988) database and/or belong to the Christakos et al. (2005) database ofmortality rates, we have identified 363 towns in which Jews were present at the onset of the BlackDeath in 1347. Of these 363 Jewish communities, we can match 124 locations to our database ofmortality rates. Our main sample thus consists of 124 towns with a Jewish community at the onsetof the Black Death (1347) and for which we know the Black Death mortality rate (%, 1347-1352).

3. Black Death Mortality RatesOur data on cumulative Black Death mortality rates (%) for 263 localities for the whole period1347-1352 is based on the estimates collected by Christakos et al. (2005) which come from a widerange of historical sources. We verify and supplement these where possible with data from othersources including Ziegler (1969), Russell (1972), Pounds (1973), Gottfried (1983), and Benedictow(2005). These localities belong to 13 countries of Western Europe using today’s boundaries:Austria, Belgium, the Czech Republic, France, Germany, Ireland, Italy, Norway, Portugal, Spain,Sweden, Switzerland and the United Kingdom. We have a percentage estimate of the mortalityrate for 166 of these 263 localities. For example, Cologne, Granada, Milan and Zurich and hadan estimated cumulative Black Death mortality rate of 35%, 30%, 15% and 60% respectively. Forthe 96 other localities, the sources report more qualitative estimates: (i) For 49 towns, Christakoset al. (2005) provide a literary description of mortality. We rank these descriptions based on thesupposed magnitude of the shock, and assign each one of them a numeric mortality rate: 5%


for “spared” or “escaped”, 10% for “partially spared” or “minimal”, 20% for “low”, 25% for“moderate”, 50% for “high”, 66% for “highly depopulated”, and 80% if the town is “close to beingdepopulated” or “decimated”; (ii) For 19 towns, we know the mortality rate of the clergy in thetown. Christakos et al. (2005, p.138) cite Ziegler (1969), who argues that “it would be reasonable tostate as a general rule that the proportion of benefited clergy who died in any given diocese couldnot possibly have been much smaller than the corresponding figure for the laity and is unlikely tohave been very much bigger. Arbitrary limits of 10% less [mortality among benefited clergy] and25% more [mortality among benefited clergy] seem to provide a reasonable bracket within whichthe correct figure must be encompassed.” This suggests that clergy mortality was on average8% higher than general mortality. We thus divide the clergy mortality rates by 1.08 to obtain themortality rate of these 19 towns; and (iii) For 29 towns, we know the desertion rate of the town,which includes both people who died and people who never came back. Christakos et al. (2005,p.154-155), using data on both desertion rates and mortality rates available for 10 towns, showthat the desertion rate is on average 1.2 times higher than the mortality rate. We thus divide thedesertion rates by 1.2 to obtain the mortality rate of these 19 towns.

4. Black Death SpreadWe use the raw data from Christakos et al. (2005) to obtain for 95 towns among the 124 townsof our main sample the year and month of first infection in the town (the day is almost neveravailable). For the other 29 towns, we rely on information for very neighboring towns in thedata and maps of the epidemic available in Christakos et al. (2005, Figures 3-4), as well as extrasources to impute the year-month of first infection. For example, for Landshut in Germany, welearn from Benedictow (2005, 190) that the epidemic went from Muhldorf to the neighboring townof Landshut (50 km). From Christakos et al. (2005), we know that Muhldorf and Regensburg werefirst infected in June and July 1349, respectively. Since Landshut is about one-half of the waybetween Muhldorf and Regensburg, it must have been infected in June or July 1349, but mostprobably in June 1349. Another example is Monthey in Switzerland, which is less than 10 Kmfrom Saint-Maurice, a town that was infected in January 1349. We thus use January 1349 forMonthey. A last example is Pamplona in Spain. We know from the raw data from Christakoset al. (2005) that Navarra was first hit in October 1348. We thus use October 1348 for Pamplona.Information is sparser for the year-month of last infection, and thus the duration of the epidemicin each town, and only available for 39 towns among the 124 towns of our main sample.

5. Missing Black Death Mortality Rates

Our full sample consists of 1,869 Western European towns that reached 1,000 inhabitants at onepoint between 800 and 1850 in the Bairoch (1988) database and/or belong to the Christakos etal. (2005) database of mortality rates. These towns belong to 18 countries: Austria, Belgium,the Czech Republic, Denmark, France, Germany, Ireland, Italy, Luxembourg, Norway, Poland,Portugal, Slovakia, Spain, Sweden, Switzerland, The Netherlands and the United Kingdom. Dataon Black Death mortality rates for the other 1,869 - 263 = 1,606 cities do not exist.


In order to extend our analysis to these other cities, we use spatial analysis to impute the missingvalues. Our assumptions in doing this are that (1) there exist some underlying causes of mortalityrates which are unobserved, (2) these causes have a large random component (i.e. are external toour model of persecution), (3) these causes are also spatially correlated. For example, it is widelyacknowledged that fleas living off of rat populations were a primary vector for the plague. It ishighly plausible that a latent variable measuring the suitability of a city’s surrounding region forsustaining large rat populations satisfies the three criteria laid out above.

Using mortality rates from neighboring towns. Using GIS, we calculate the Euclidean distancebetween each town and each other town in our sample of 1,869 towns. For the towns for whichmortality is missing, we proxy their mortality rate by the mortality rate of the closest neighboringtown with mortality data and if this town is within 10 km or 50 km.

Figure A.2: Predicted vs. Measured Mortality Rates at the Optimal Power.

Using spatially extrapolated mortality rates. We create a two-dimensional surface of predictedplague mortality using an inverse distance weighted function of known mortality rates. For everypoint on the surface a predicted mortality rate is generated using the closest 15 cities withinan approximately 1,000 km radius circle around the point. For a point on the surface, x, withunknown mortality the influence of city, i, with known mortality diminishes with its distance fromx according to the weights used. These weights are determined by a parameter, p � 0, referredto as power. As the power decreases, the influence of more distant points increases. If p = 0, thenall points receive equal weight in determining all other points on the map. The influence of moredistant points decreases exponentially as p increases. To create our mortality estimates we choosean optimal p using cross-validation techniques. The procedure begins by choosing some power, p.Then, using the sample of n localities with known mortality rates, we create a predicted mortalityrate surface using all of the cities except for city j. We then predict the mortality rate for city j


as mj using our mortality surface and create it’s residual as (mj � mj) where mj is the knownmortality rate for city j. This procedure is then repeated to create predicted mortality rates andresiduals for all the remaining cities. We then calculate the Root Mean Square Error (RMSE) of the

residuals created as RMSE(p) =qPn

i=1(mi�mi)2

n , where i indexes the cities with known mortalityrates. This procedure is repeated for a large number of choices of p and then the optimal power ischosen as the one which minimizes the RMSE.

We generate optimal mortality surfaces using the 263 localities for which we know the mortalityrate. The cross-validation exercise chooses an optimal power for creating the mortality surfaceas 1.0136. Web Appx. Fig. A.2 shows the measured versus predicted values using the optimalpower. Web Appendix Figure A.3 shows the location of the 263 localities for which we know thenon-extrapolated mortality rate and the optimal mortality surfaces with the extrapolated rates.

Figure A.3: Extrapolated Black Death Mortality for Western Europe.

!

!

!

!

!!

! !!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

! !

!

!

!

!!

!

!!

!

!

!

!!

!

!!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!!

!

!

!

!

!

!

!

!

!!

!

!

!

!

!

!

!

!

!

!!

!!!

!

!!

!

!

!

!!

!

!!

!!

!

!

!

!

!

! !

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!

!!

!!

!

! !

!

!

!

!

!

!

!

!

!

!

!

!!

!

!!

!

!

!

!

!

! !

!!

!

!

!

!

!

Predicted Mortality (1.5)0 - 2021 - 4041 - 6061 - 8081 - 100

! Cities With Mortality Data

Using mortality rates from provinces. Christakos et al. (2005) report cumulative mortality ratesfor selected provinces (e.g., “Languedoc”). Using information from Wikipedia, we assign eachtown to the province it historically belonged to and attribute to each town with missing mortalitydata the mortality rate of its province. We obtain a mortality estimate for 35 towns, using Aragon(7 towns), Bohemia (5), Castile (4), Catalonia (2), Duero Valley (3), Hesse (5), Languedoc (1),Mallorca (1), Navarra (1), Palencia (1), Pomerania (1), Savoy (1) and Sicily (3).

6. Jewish Presence and Jewish PersecutionThe main source of information for Jewish presence and Jewish persecutions and expulsions isthe Encyclopedia Judaica (Berenbaum and Skolnik, 2007). We supplement the Encyclopedia Judaicausing a variety of sources which we detail below. The Encyclopedia Judaica is a multi-volume


English language encyclopedia of the history of the Jewish people. We use the most recent andcomprehensive 2007 edition which comprises 22 volumes in print as is available online.

City-Level Entries. An example entry of the Encyclopedia Judaica reads as follows:

In the 13th and 14th centuries the Krems community was one of the most importantin Austria. The Jews were moneylenders and they were not restricted to dwellings inany one quarter of the city. Persecutions occurred in 1337 and 1347. On Sept. 29, 1349,inflamed by rumors that the Jews had caused theBlack Death, the populace of Kremsand the nearby villages massacred most of the Jews and plundered their homes. A fewescaped to the fortress. Duke Albrecht V ordered his soldiers to punish the attackers,laid penalties on the city, and sentenced three of the ringleaders to death. In 1355 Jewsare recorded as living in Krems, owning houses all over the city,

This entry provides information about the presence of a Jewish community, the economic role theyplayed in the city, what restrictions on residency they faced and details about persecutions bothbefore and during the Black Death. The entry also makes it clear that the Jews were scapegoatedfor the Black Death and that it was the mob who were responsible for the massacre while the localruler attempted to protect them.

The entry for Toulon is as follows:

TOULON, port in the Var department, S.E. France. In the second half of the13thcentury the Jews made up an appreciable proportion of the population of Toulon:at a general municipal assembly held in 1285, 11 of the 155 participants were Jews.They shared the same rights and duties as the other citizens. The community came toa brutal end on the night of April 12/13, 1348 (Palm Sunday), when the Jewish street,“Carriera de la Juteria,” was attacked, the houses pillaged, and 40 Jews slain; this attackwas probably related to theBlack Deathpersecutions. Faced with an enquiry set up bya judge from Hyeres, the assailants fled; however, they were soon pardoned. Afterthis date, in addition to a few converted Jews, there were in Toulon only individualJews who stayed for short periods; one such man was Vitalis of Marseilles, who wasengaged as a town physician in 1440.

Other entries in Encyclopedia Judaica do not provide as much information. In many cases we crossreferenced the data using a host of different sources. The most important supplementary sourceswe use are Beinart (1992). In particular, the detailed maps provided by Beinart (1992) provideinformation on a number of communities that are not mentioned by the Encyclopedia Judaica.

For every entry in the Encyclopedia Judaica we collect information on the date at which Jewswere first mentioned in a city. For entries where the entry date was unclear or unrecorded weassume that a Jewish community had been present for the 50 years (two generations) or 25 years(one generation) prior to their first mention in an entry, depending on the information available.We recorded information on every subsequent entry to or exit from a city mentioned in theEncyclopedia and collected data on every single persecution for the period up to 1600.


Region-Level and Country-Level Entries. To supplement the individual city/community entriesin Encyclopedia Judaica we also collect information from the country and regional level entries. Forexample, the entry for Alsace contains the following information:

The first evidence of Jews in Alsace is reported by Benjamin of Tudela who mentions(c. 1170) Jews in Strasbourg. From the beginning of the 13th century, Jews are alsomentioned in Haguenau, Obernai, and Rosheim, and later, during the same century,in Wissembourg, Guebwiller, Colmar, Marmoutier, Rouffach, Ensisheim, Molsheim,Mulhouse, and Thann. Probably many refugees expelled from France in 1306 wentto Alsace. Jews are henceforward found residing in some 40 additional localitiesthere, notably, Ribeauville, Selestat, Bouxwiller, Kaysersberg, and Saverne. The Jewsof some 20 communities in Alsace were victims of the Armleder massacres, principallyat the beginning of 1338. Further anti-Jewish persecutions affected the communities ofColmar, Selestat, Obernai, Rosheim, Mulhouse, Kaysersberg, Turckheim, and Munsterin 1347. Later, the Jews were accused of spreading the Black Death , even before theepidemic began to ravage Alsace. A gathering of nobles and representatives of theimperial cities of Alsace discussed the subject in Benfeld at the beginning of 1349,and the city of Strasbourg alone defended the Jews. Subsequently, the Jews werecruelly put to death in some 30 towns in Alsace. After the artisans gained control ofthe municipal council of Strasbourg, having eliminated the patricians, the importantJewish community of this city met the same fate. These events left their mark on thefolklore and the toponyms of Alsace. The Jews reappeared in several towns of Alsaceafter a short while, apparently with an improved legal status.

Other sources. Due to possible concerns about the depth of coverage in the Encyclopedia Judaicawe supplemented this core dataset with information from the Jewish Encyclopedia published in1906 (Adler and Singer, eds, 1906) and a range of specialized studies for different cities, regionsand countries. These include Abulafia (2002); Baron (1965a,b, 1967a,b); Benbassa (1999); Botticini(1997); Chazan (2006); Cluse, ed (2004); Emery (1959); Foa (2000); Golb (1998); Jordan (1989);Jordon (1997); Kisch (1949); Klein (2006); Levenson (2012); Nahon (2002); Quesada (2004); Ries(1995); Roth (1950, 1961, 2014); Schwarzfuchs (1967); Shank (1988); Spector and Wigoder, eds(2001); Segre (1986); Shatzmiller (1974); der Pfalz, ed (2005); Taitz (1994); Toch, ed (2003). Whenusing such sources, there are 12 observations where we cannot be entirely certain that there wasa Jewish community intact in 1347. These 12 observations include 8 French towns for whichinformation is sparse following the general expulsions of 1306, 1315 and 1322 (since the expulsionsonly concerned cities belonging to the Kingdom of France, whose territory covers much less thanFrance’s territory today), as well as 3 Tuscan towns in which Jews settled in the 14th century butfor which we cannot be sure of the year. There are also 20 observations where we cannot be entirelycertain about the fate of the community during the Black Death period. The 20 observationsinclude towns for which one or several sources mentioned a persecution during one of the BlackDeath years, but without providing corroborating details about it.


7. Main Control Variables

Average Growing Season Temperature 1500-1600. We use temperature data from Luterbacheret al. (2004). They reconstruct seasonal European temperatures (celsius degrees) since 1500 usingproxy data from ice cores, tree rings, and written records. The data cover 0.5� x 0.5� grids whichis approximately 50km x 50 km at European latitudes. The data extend from 25� W to 40� E and35� N to 70� N which includes all of the cities in our full sample. We extract the growing season(summer) temperature for each of our cities during the 16th century as this is the closest centuryto the Black Death period for which we have data. No comparable data exist for earlier centuries.

Elevation. City elevation data come from Jarvis et al. (2008) which is available at http://srtm.csi.cgiar.org. This data reports elevation in meters. The spatial resolution between 1 and 3 arc-seconds. Where there is missing data we have supplemented it using Wikipedia.

Cereal Suitability. Our data are from the FAO Global Agro-Ecological Zones (GAEZ) dataset asdescribed in Fischer et al. (2002). We use these in preference to the Ramankutty et al. (2002) asthe latter does not have full coverage for all of Western Europe. We use the GAEZ’s overall cerealsuitability data assuming low inputs and rain-fed irrigation. The data are available for a resolutionof 5 arc minute cells, or approximately 10 km X 10 km at the equator. We then extract the averagesoil suitability for the closest cell to the city. Overall cereal suitability is scaled from 1-9 where 1 isbest, 8 is unsuitable and 9 is water (seas and oceans are treated as missing values).

Pastoral Suitability. We control for the potential suitability of a region surrounding a city forpastoral farming with a variable measuring grazing suitability. This variable come from Erb et al.(2007) who create land use measures at a resolution of 5 arc minute cells, or approximately 10 kmX 10 km at the equator. This records how land is used in each cell in 2000. The five categoriesthey code for are: cropland, grazing, forestry, urban, and areas without land use. Their grazingcategory is calculated as a residual after accounting for the percentage of area taken up by the otherfour uses. As part of this analysis, they also generate a variable measuring the suitability of eachcell for grazing (as opposed to actual present-day use). The suitability measure is created by firstseparating grazing land into three categories based on cover: “high suitability of cultivated andmanaged areas, medium suitability of grazing land found under tree cover, and low suitability ifshrub cover or sparse vegetation is detected in remote sensing” (Erb et al., 2007, 199). They thenfurther subdivide the first two of these categories into areas with a net primary productivity ofCarbon per meter squared is greater than 200 grams and those in which it is less than 200 grams.This results in five categories which they regroup into four categories with 1 = most suitable and4 = least suitable. There is a fifth category which is “no grazing” which we re-code as 5. We thenextract the average soil suitability for the closest cell to the city.

Distance to the Coast and Major Rivers. We create a variable to measure distance to the coast andmajor rivers in meters using ArcGIS. We base these distances on the 1300 shape file downloadedfrom Nussli (2011). We verify the information for some cities using Google Earth.

Town Population Estimates. Our main source of urban population data is Jebwab et al. (2016),

http://srtm.csi.cgiar.org

http://srtm.csi.cgiar.org


who combine data from both Bairoch (1988) and Chandler (1987). The last two sources representattempts to collect information on population for all towns with at least 1,000 inhabitants. Moreprecisely, the Bairoch (1988) dataset of city populations. The Bairoch dataset reports estimatesfor 1,797 cities between 800 and 1850. This dataset has been widely used by a range of scholarsstudying premodern urbanization and economic development. We follow Bosker et al. (2013) andVoigtlander and Voth (2013) in updating the Bairoch dataset where a consensus of historians haveprovided revised estimates of the population of a particular city, including Bruges, Paris, andLondon. We supplement Bairoch with Chandler (1987) which is in some cases more specific inthe sources used to measure city population. For the 1,869 - 1,797 = 72 towns that are not in theextended Bairoch (1988) data set but belong to the Christakos et al. (2005) data set, we sometimesuse for the year 1300 information on pre-plague population available in Christakos et al. (2005).For the town-year observations for which population is still not available, we believe that it mustbe less than 1,000 and thus arbitrarily assume that their population was 500.

Roman Romans. Data on Roman roads is provided by the Digital Atlas of Roman and MedievalCivilizations: http://darmc.harvard.edu/icb/icb.do?keyword=k40248&pageid=icb.page601659.We use this shape file to create two distances: (1) distance to all Roman roads and (2) distance to“major” Roman roads. Since major settlements often formed along the intersections of the Romanroad network, we also use GIS to create a variable for (1) distance to all Roman road intersectionsand (2) distance to major Roman road intersections. We then create a dummy variable that takesthe value of 1 if a city is within 10 kilometers of a road or an intersection.

Medieval Trade Routes. We use Shepherd (1923) to obtain the path of major medieval land traderoutes. We use ArcGIS to create a shape file that allows us to measure distance to medieval routesor the intersection of two major trade routes. We create a dummy variable that takes the value of1 if a city is within 10 kilometers of a medieval trade route or an intersection.

Medieval Fairs. We obtain data on the location of important medieval fairs from two sources.The first source is Shepherd (1923). The second source is the Digital Atlas of Roman and MedievalCivilizations: http://darmc.harvard.edu/icb/icb.do?keyword=k40248&pageid=icb.page188865.Using this information, we create a dummy equal to one if there was a fair in the town.

Hanseatic League. We document whether or not a city was a member of the Hanseatic League.We do this by matching where possible the extended Bairoch city data with available lists of citieswhich belonged to the League. We include only cities which were members of the League and donot include cities with Hansa trading posts or Hansa communities. Our main source is Dollinger(1970). Using this information, we then create a dummy if the town belonged to the League.

Market Access to All 1,869 Towns. Following Donaldson (2017), we calculate market access forcity j as MAjt =

Pi 6=j Nit⌧

��ji , where city i is any other city in the full sample of 1,869 cities and

Nit is the time-varying measure of city i’s population. We calculate market access by assigning theleast cost itinerary across cells of 10x10 km between city i and city j. To measure the ⌧ we employdata on travel speeds as this reflects the time it took for merchants or other travels to move from

http://darmc.harvard.edu/icb/icb.do?keyword=k40248&pageid=icb.page601659

http://darmc.harvard.edu/icb/icb.do?keyword=k40248&pageid=icb.page188865


city to another (and hence potentially for infected rats and fleas to be transported from one city toanother) (see Boerner and Severgnini, 2014). We normalize all travel speeds to the most expensiveform of transport (portage). Our benchmark travel speeds are from Boerner and Severgnini (2014)and based on discussions in Pryor (1992); McCormick (2001). Normalizing the speed to portersto 1, this assigns a travel cost of 0.5 to road and river transport and 0.18 to sea transport. Seatransport and road/river transport were thus about 5 and 2 times faster than portage respectively.Lastly, in our benchmark analysis we set � = 3.8 (also from (Donaldson, 2017)).

Aqueducts. We use GIS to create a shape file for whether or not a town was within 10 km froma Roman aqueduct using the map provided by Talbert, ed (2000) as well as information from twoWikipedia webpages: https://en.wikipedia.org/wiki/List of aqueducts in the Roman Empireand https://fr.wikipedia.org/wiki/Liste des aqueducs romains.

Medieval Universities. Bosker et al. (2013) provide data on the presence of medieval universitiesfor European cities with populations greater than 10,000 (at some point between 800 and 1800).We consulted Wikipedia and other sources to find evidence of medieval universities in Europeancities with smaller populations. There are five medieval universities missing from the list in Boskeret al. (2013): Angers, Greifswald, Ingolstadt, Tuebingen, and Uppsala. However, as none of theseuniversities were established prior to the Black Death, we do not include them in our analysis.

Monarchy in 1300. We construct information on whether or not a city was ruled by a majorkingdom using the shape files provided by Nussli (2011) who report political boundaries inEurope for every century. We then assign each city to its political boundary in 1300 by hand.We assign a city as belonging to a monarchy in 1300 if it belonged to the Kingdom of Bohemia, theKingdom of Denmark, the Crown of Castile, the Kingdom of France, the Kingdom of Norway, theKingdom of England, the Kingdom of Sicily in Naples, the Kingdom of Granada, the Kingdom ofScotland, the Kingdom of Hungary, the Kingdom of Sicily, the Kingdom of Galicia-Volhynia, theCrown of Aragon, the Kingdom of Portugal, the Kingdom of Majorca, the Kingdom of Sweden.

State Capital in 1300. We use the data provided by Bosker et al. (2013) who collect data on capitalcities from McEvedy and Jones (1978).

Parliamentary Activity and Distance to Parliament. Our data on parliamentary activity is fromZanden et al. (2012). This measures the number of times that Parliaments met at a regional levelin 1300–1400. We create a dummy variable based on whether or not a town is in a region/countrywhich had above the median number of parliamentary meetings. We also obtain a list of whetherthe parliaments were held for each region/country. We then use GIS to compute for each city theminimal Euclidean distance to a parliament.

Self-governing City. Bosker et al. (2013) provide information on the existence of communes fora subset of the cities in the Bairoch dataset. Bosker et al. (2013) create a variable“commune” thattakes a value of 1 if there is indication of the presence of a local urban participative organizationthat decided on local urban affairs. Stasavage (2014) provided us with data on 169 cities thatwhere autonomous at some point between 1000 and 1800. We utilize the variable for 1300-1400.

https://en.wikipedia.org/wiki/List_of_aqueducts_in_the_Roman_Empire

https://fr.wikipedia.org/wiki/Liste_des_aqueducs_romains


Stasavage (2014) defines autonomous cities in the following terms:

I have defined an “autonomous city” as being one in which there is clear evidence thatsuch institutions of self-governance existed, and in addition there is also clear evidenceof exercise of prerogatives in at least one of the policy areas referred to above. In theabsence of such evidence the default is to code a city as non-autonomous (6).

As Stasavage (2014) notes, his definition of city autonomy is stricter than the definition ofcommune used by Bosker et al. (2013). We create a dummy equal to one if the city is a communein the Bosker et al. (2013) data set or a self-governing city according to Stasavage (2014).

Battles. As our main source we use Wikipedia’s list of all battles that took place between 1300 and1350. https://en.wikipedia.org/wiki/List of battles 1301-1800. This is a highly reliable sourcefor the most important battles of the period. We are not concerned about sample selection here asWikipedia’s coverage of European history is extensive; battles not listed on Wikipedia are likely tohave been extremely small. For each battle we assign a geo-coordinate based on either the locationof the battle or the location of the nearest town or city mentioned in the entry. We exclude navalbattles and conflicts which cannot be located (such battles were typically extremely minor).

8. Investigation of Causality

Density. Data on walled density (population divided by walled area) for 56 towns comes from(Cesaretti et al., 2016). Web Appendix Figure A.4 shows there is no relationship between mortalityand log walled density for these 56 towns (mort. = 27.85** + 2.12 log.wall.dens.; R2 = 0.01).

Figure A.4: Black Death Mortality and Walled Density 1300.

Panel. Since the Black Death lasted 6 years, we create panel data for 10 six-year periods t bothbefore 1347 (1297-1302, . . . , 1341-1346) and after 1352 (1353-1358, ..., 1398-1403). For 172 towns i

that had Jews at one point in 1297-1403 and for which we know the mortality rate, we regress adummy if there was a persecution in town i in period t on the mortality rate in the same town

https://en.wikipedia.org/wiki/List_of_battles_1301-1800


and period, for the town-periods where Jews were present. We include town fixed effects andperiod fixed effects, and posit the mortality rate is equal to 0 outside 1347-1352. Column 1 of WebAppendix Table A.1 shows that the baseline effect remains unchanged when doing so. Column 2shows that this result holds even if we drop the ten post-1352 periods, since there have been plaguereoccurrences after the Black Death and consistent data does not exist on the specific mortality rateassociated with each reoccurrence, so mortality could be mismeasured.

Table A.1: MORTALITY RATES AND PERSECUTIONS, PANEL (1297-1403)

Dependent Variable: Dummy if Any Jewish Persecution in Town i in Period t:

Period: (1) 1297-1403 (2) 1297-1352

1. Effect of Mortality Rate Town i Period t -0.009*** [0.002] -0.009*** [0.002]

Town-Period Observations 2,581 1,346Town Fixed Effects, Period Fixed Effects Yes Yes

Notes: Robust SE’s clustered at the town level: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.

Spatial Fixed Effects. Our towns belonged to many small states circa 1347, and we do not haveenough observations to be able to include historical state fixed effects or modern country fixedeffects. However, rows 1-3 of Web Appendix Table A.2 show that the baseline effect remainssignificantly negative when including a dummy for the Holy Roman Empire (row 1), threecultural areas fixed effects (row 2; Southern Europe, Western Europe and Central Europe; source:Wikipedia (2018) article “Cultural Area”) or three linguistic areas fixed effects (row 3; “latin”,“germanic” and “slavic”; source: Wikipedia (2018) article “Linguistic regions of Europe”).

Table A.2: MORTALITY AND PERSECUTIONS, SPATIAL FIXED EFFECTS



1. Including Holy Roman Empire Fixed Effect (N = 1) -0.008*** [0.002] 0.659*** [0.126] 1242. Including Cultural Area Fixed Effects (N = 3) -0.005*** [0.002] 0.681*** [0.098] 1243. Including Linguistic Area Fixed Effects (N = 3) -0.005** [0.002] 0.651*** [0.097] 124


Size of the Jewish Community. The majority of entries in the Encyclopedia Judaica do not containinformation on the size of Jewish communities. The information that is available suggests that ingeneral these communities were small. Using the Encyclopedia Judaica and the Jewish Encyclopedia,as well as various historical sources, we obtain for 30 towns in our main sample of 124 towns thepercentage share of Jews in the town circa the Black Death or the decades just before the BlackDeath. In some cases, the source directly gives this percentage share. For example, in Aix weare told that they were estimated to have been 9.1% of the population in 1341. In other cases, weknow the population size of the Jewish community. For example, we learn that the community ofMontpellier had 2,250 residents — 6.4% of the population — just before the Black Death.

Web Appendix Table A.3 show that results are also unchanged if we limit our sample to towns


Table A.3: JEWISH COMMUNITY SIZE, MORTALITY RATES AND PERSECUTIONS



1. Keep if Jewish Cemetery, Quarter or Synagogue -0.009** [0.004] 0.841*** [0.163] 592. Keep if First Year of Entry in the Town Median (= 1155) -0.008** [0.004] 0.694*** [0.164] 623. Keep if Last Year of Entry in the Town Median (= 1200) -0.009*** [0.003] 0.831* [0.140] 674. Keep if Jewish Centrality Index � Median (= 20.7%) -0.011*** [0.003] 0.981* [0.130] 62

5. Drop if Jewish Cemetery, Quarter or Synagogue -0.009*** [0.003] 0.822*** [0.154] 656. Drop if First Year of Entry in the Town Median (= 1155) -0.010*** [0.003] 0.960*** [0.130] 627. Drop if Last Year of Entry in the Town Median (= 1200) -0.009*** [0.003] 0.833* [0.157] 578. Drop if Jewish Centrality Index � Median (= 20.7%) -0.008*** [0.003] 0.713*** [0.148] 62

9. Drop if Year of Last Entry During 1347-1352 (Parchim in 1350) -0.009*** [0.002] 0.827*** [0.103] 123


where the community was likely large or small, i.e. long before 1347. In particular, we keep ordrop the cities, (i) with a Jewish cemetery, quarter or synagogue (rows 1 and 5), (ii) whose year offirst entry or last reentry in the town was below the median in the sample (rows 2-3 and 6-7), and(iii) whose Jewish centrality index was above the median in the sample (rows 4 and 8).

Note that we calculate our measure of the network centrality of a Jewish community using ourdata on Jewish presence from Berenbaum and Skolnik, eds (2007). For town i, and other towns j 2J (363 towns with a Jewish community circa 1347) or j 2 A (all 1,869 towns), the Jewish centralityindex is equal to

Pj2J D

��ij ÷

Pj2AD

��ij ⇤ 100 with Dij the travel time between city i and city j.

If all surrounding towns have a Jewish community, it will be close to 100, and 0 otherwise. WebAppendix Figure A.5 depicts this measure of Jewish centrality.

Figure A.5: Jewish Centrality Index for the 124 Towns of our Main Sample.

In Web Appendix Table A.4, we show that mortality was not correlated with the population shareof Jews and the log of their population (column (1)), for the 30 towns among our main sample of


124 towns and for which we have data on the size of the community. We then show that mortalityis not correlated with various proxies for the size of the community, for our main sample of 124towns: three dummies equal to one if the town had a Jewish cemetery, a Jewish quarter or aJewish synagogue (column (2)), the year of first entry ever and the year of last reentry in the town(column (3)), the Jewish centrality index (column (4)), and two dummies equal to one if there wasa persecution in the town in 1321-1346 or 1300-1346 (column (5)). We simultaneously test all thesecontrols in column (6). Likewise, for a sample of 172 cities with mortality data and that do notbelong to the British Isles and Scandinavia, we find no correlation between mortality and whetherthere was a Jewish community in the town (column (7)). Note that we exclude the British Islesand Scandinavia because there was a blanket ban on the presence of Jews for their whole territory.

Table A.4: INTENSIVE AND EXTENSIVE MARGIN OF JEWISH PRESENCE AND MORTALITY

Dependent Variable Black Death Mortality Rate (%, 1347-1352):

(1) (2) (3) (4) (5) (6) (7)

Population Share of Jews (%) 0.29 0.27[0.33] [0.46]

Log Number of Jews 3.34 3.62[3.78] [4.77]

Dummy if Jewish Cemetery 1.12 2.52[4.48] [8.61]

Dummy if Jewish Quarter -2.27 4.88[4.14] [9.45]

Dummy if Jewish Synagogue -1.06 -3.09[3.90] [11.82]

Year of First Entry 0.00 0.02[0.00] [0.06]

Year of Last Entry -0.00 -0.03[0.01] [0.06]

Jewish Centrality Index 0.01 -0.10[0.05] [0.13]

Dummy if Persecution 1321-1346 -5.06 -34.79[8.01] [25.96]

Dummy if Persecution 1300-1346 0.48 26.53[5.69] [23.59]

Dummy if Jews Present 1347-1352 -2.80[3.30]

Observations 30 124 124 124 124 30 172


Endogeneity of Jewish Presence. One could worry that Jewish presence circa 1347 is endogenous.If Jewish presence is better measured in towns with high mortality and low persecution or lowmortality and high persecution, results are not valid. In our sample, there is only one town whenJews came after 1347, Parchim, in 1350. The Black Death broke out there after Jews came. Row 9of Web Appendix Table A.3 shows results hold if we drop that city, to only rely on towns whoseJewish presence was predetermined to the Black Death (1347). Results are also unchanged if we


limit our sample to towns where Jews were historically present, i.e. long before 1347. In particular,we keep the cities, (i) with a Jewish cemetery, quarter or synagogue (row 1), (ii) whose year of firstentry or last reentry in the town was below the median in the sample (rows 2-3), and (iii) whoseJewish centrality index was above the median in the sample (row 4). These tests allow us to onlykeep the cities and regions where Jews were likely to be well-established.

Number of Victims For the vast majority of towns we do not know either the size of the Jewishcommunity or the number of victims. However, information on the number of victims is availablefor a limited number of cities. We consulted the Berenbaum and Skolnik, eds (2007) and a rangeof other sources including Adler and Singer, eds (1906) to collect what information is available onthe number of victims for some cities. For example, in Brussels, 600 Jews are said to have beenkilled. In Grenoble the number given is 74, while 40 Jews are said to have died in Erfurt.

Table A.5: FIRST STAGE OF THE MAIN INSTRUMENTAL VARIABLES REGRESSIONS

Dependent Variable: Black Death Mortality Rate (%, 1347-1352):

IV Strategy: IV1: Proximity to Messina IV2: Month of First InfectionTable 4 in the Paper: Row 2 Row 3 Row 7 Row 8

Log MA to Messina 4.42*** 3.46**[0.79] [1.67]

#Months between Oct 1347 and First Infection -0.87*** -0.83***[0.15] [0.31]

Log MA to All 1,869 Towns -1.75* -0.32[0.94] [0.93]

Longitude -1.54** -0.87**[0.62] [0.41]

Latitude 8.23 -1.94[6.94] [7.88]

Longitude Squared 0.13*** 0.13***[0.04] [0.03]

Latitude Squared -0.09 0.02[0.08] [0.09]

F-stat 31.0 4.3 33.3 7.32Observations 123 123 124 124


IV1: Proximity to Messina. For the 124 towns, we construct an instrumental variable based on atown’s log market access to Messina, conditional on a town’s log market access to all 1,869 towns.Market access to Messina m for town i is defined as MAim = ⌃(Lm ÷ ⌧

�im), with Lm being the

population of Messina in 1300, ⌧im the computed travel time between town i and Messina, and� = 3.8. To compute the travel times, we use the travel speeds from Boerner and Severgnini (2014),the same way we did for market access to all 1,869 towns (see paragraph Market Access to All 1,869Towns in Section 7.). The first stage is shown in Web Appendix Table A.5.

IV2: Month of First Infection. We use as our second instrument the number of months betweenOctober 1347 and the month of first infection in the town. We describe how we obtain the monthof first infection in Section 4.. The first stage is shown in Web Appendix Table A.5.


IV1 & IV2: Non-Correlation with Past Town Population Growth. In Web Appendix Table A.6,we verify that in our sample of 124 towns the IVs — log market access to Messina (IV1, see column(1)) and number of months between October 1347 and the month of first infection in the town (IV2,see column (2)) — are not correlated with the log growth rate of town population in 1200-1300,conditional on log market access to all 1,869 towns and log initial population in 1200.

Table A.6: INSTRUMENTAL VARIABLES AND TOWN GROWTH IN PREVIOUS CENTURY

Dependent Variable: Log Change in Town Population in 1200-1300:

IV Strategy: (1) IV1: Proximity to Messina (2) IV2: Month of First Infection

Log MA to Messina -0.051 [0.049]#Months between Oct 1347 and First Infection 0.014 [0.011]Log MA to All 1,869 Towns 0.130** [0.059] 0.127** [0.055]Log Town Population in 1200 -0.329*** [0.053] -0.323*** [0.053]

Observations 123 124

Notes: For our main sample of 124 towns, we regress the log change in town population in 1200-1300 (i.e., log town population in1300 - log town population in 1200) on: Column (1): Log market access to Messina in 1300 (IV1); and Column (2): Number of monthsbetween October 1347 and the month of first infection (IV2). We simultaneously control for log market access to all 1,869 towns andlog initial town population in 1200. Robust SE’s: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.

IV1: Other Controls. In Web Appendix Table A.7, we show the results of the first IV strategy— using log market access to Messina as an IV, conditional on market access to all towns —hold if we: (i) Control for proxies for community size and past persecutions (cemetery, quarter,synagogue, years of first and last entry, centrality index, and dummies if persecution in 1321-1346 and 1300-1346) (rows 1-2); (ii) Control for log market access to Genoa (rows 3-4). Kaffa wasa trading colony established by Genoa on the Black Sea. The ships carrying the plague fromKaffa were bound to Genoa, and stopped in Messina because sailors were sick. By additionallycontrolling for log market access to Genoa, we capture the fact that the North-West of Italy was oneof the wealthiest areas of Europe then, and we exploit the fact that the ships stopped in Messinaand not a different city; and (iii) Control for log market access to the main cities of the Middle-East and North Africa (MENA) (rows 5-6). The cities j that we consider to construct this marketaccess variable were the largest MENA cities in 1300 according to Chandler (1987): Cairo (450,000)whose port was Damietta, Damietta itself (90,000), Fez (200,000) and Marrakech (100,000) whoseport was Ceuta, and Istanbul (100,000) and Tunis (75,000) which were their own port. To obtainthe travel times, we compute the least cost travel paths to the ports of Damietta, Ceuta, Istanbuland Tunis, respectively, via four transportation modes by sea, by river, by road and by walkingwith the transportation speeds from Boerner and Severgnini (2014). In rows 7-8, the instrumentis log distance to Messina, conditional on average log distance to all towns. In rows 9-10, it is logmarket access to Messina, conditional on log market access to all towns except Messina.

Plague Reoccurrences. Using the data from Schmid et al. (2015) based on Biraben (1975), we studythe relationship between plague reoccurrence and the persecution of Jews. We focus on 415 towns


Table A.7: ROBUSTNESS CHECKS FOR THE IV1 STRATEGY



1. IV1: Log MA Messina, Ctrls Log MA & Size & Pers. (F: 36.0) -0.016*** [0.005] 1.253*** [0.334] 1232. Row 1 + Latitude, Longitude and their Squares (F: 4.6) -0.024* [0.012] 12.015** [4.738] 123

3. IV1: Log MA Messina, Ctrls Log MA & Log MA Genoa (F: 19.2) -0.016*** [0.004] 1.185*** [0.284] 1224. Row 1 + Latitude, Longitude and their Squares (F: 4.9) -0.020* [0.012] 13.866** [6.626] 122

5. IV1: Log MA Messina, Ctrls Log MA & Log MA MENA (F: 23.1) -0.016*** [0.006] 1.1275*** [0.178] 1236. Row 3 + Latitude, Longitude and their Squares (F: 4.7) -0.020* [0.011] 5.353 [4.922] 123

7. IV1: Log Dist. Messina, Ctrl for Avg. Log Dist. All Towns (F: 7.2) -0.028** [0.014] -22.190 [17.298] 1238. Row 5 + Latitude, Longitude and their Squares (F: 5.6) -0.027** [0.011] -21.377 [16.037] 123

9. IV1: Log MA Messina, Ctrl Log MA All Towns Excl. Messina (F: 31.7) -0.021*** [0.005] 1.243*** [0.220] 1231. Row 8 + Latitude, Longitude and their Squares (F: 5.1) -0.036** [0.018] 6.351 [6.556] 123

Notes: In Rows 3-4 we instrument by log market access to Messina, controlling for log market access to all 1,869 towns (IV1) and logmarket access to Genoa. Rows 5-6: Instrumenting by log market access to Messina, controlling for log market access to all 1,869 towns(IV1) and log market access to Middle-East and North Africa (MENA). Market access for city i is defined as MAi = ⌃jPj/D�

ij , withPj being the population of town j 6= i, Dij the travel time between city i and city j, and � = 3.8. The cities j that we consider were thelargest cities of the MENA region in 1300 according to Chandler (1987): Cairo (450,000) whose port to Europe was Damietta, Damiettaitself (90,000), Fez (200,000) and Marrakech (100,000) whose port was Ceuta, Istanbul (100,000) which was its own port, and Tunis(75,000) which was also its own port. To obtain the travel times, we compute the least cost travel paths to the ports of Damietta, Ceuta,Istanbul and Tunis, respectively, via four transportation modes — by sea, by river, by road and by walking — with the transportationspeeds from Boerner & Severgnini (2014). Robust SE’s: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.

i with Jews at one point in 1353-1598. Since the Black Death lasted 6 years, we create 41 six-yearperiods t (1353-1358, . . . , 1594-1598). We then regress a dummy if there was a persecution in townin period t on the number of years with a plague outbreak within 5 km, or 100 km, from the townduring the period, for the town-periods where Jews were present. We include town fixed effectsand period fixed effects. Note that we also use outbreaks within 100 km because Biraben recordedplague outbreaks for large cities only (Roosen and Curtis, 2018), hence the need to rely on localbuffers around these. There is also no data on the mortality rate associated with each outbreak,so our intensity measure is very imperfect. Nonetheless, Web Appendix Table A.8 shows thereis a negative effect on plague reoccurrences on persecution probability, which is however notsignificant when using 5 km (column 1) but significant when using 100 km (column 2).

Table A.8: PLAGUE REOCCURRENCE INTENSITY AND PERSECUTIONS, PANEL (1353-1598)

Dependent Variable: Dummy if Any Jewish Persecution in Town i in Period t:

Reoccurrence Dummy Defined Using Plague Outbreaks: (1) Within 5 Km (2) Within 100 Km

1. Effect of Plague Reoccurences Town i Period t -0.006 [0.005] -0.005*** [0.002]

Town-Period Observations 13,187 13,187Town Fixed Effects, Period Fixed Effects Yes Yes

Notes: Robust SE’s clustered at the town level: * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.


9. Preventive Persecutions

Distance to Chillon Castle, Towns Warned of a Conspiracy, the Rhine River, and the Path of theFlagellants. We use GIS to calculate the Euclidean distance to: (i) the origins of the well poisoninglibel (the town of Chillon, whose geographical coordinates we obtain from Wikipedia); (ii) the first10 towns to be warned by letter of a Jewish conspiracy (the towns of Chambery, Geneva, Lausanne,Bern, Solothurn, Zofingen, Basel, Feiburg, Strasbourg and Koeln, whose geographical coordinateswe obtain from our data set or Wikipedia); (iii) the Rhine Towns (all the towns within 10 Km froma GIS file of the Rhine that we obtain from http://www.eea.europa.eu/data-and-maps/data/wise-large-rivers-and-large-lakes) along which the well poisoning libel spread; and (iv) the pathof the flagellants, which we recreate in GIS based on the map in (Barnavi, 1992).

Date of the Persecution. We know the year of all persecutions. Since we know the year offirst infection, we can identify persecutions that were likely preventive (if the year of persecutionappears to be the year before the year of infection) or possibly preventive (if the year of persecutionis the same year as the year of first infection) based on the year. We then know the year and monthof persecution for many towns. Since we know the year-month of first infection, we can identifypersecutions that were likely preventive (if the year-month of persecution appears to be before theyear-month of infection) or possibly preventive (if the year-month of persecution is close to theyear-month of first infection) based on the year and month.

10. Alternative OutcomesPogroms vs. Expulsions. In our main sample of 124 cities, there were 53 pogroms and 13expulsions. In 8 cases, communities suffered both a pogrom and an expulsion. For example inBerlin: “In 1349, the Jews were accused of starting the Black Plague that was sweeping throughEurope, and were expelled but not before many were killed, and had their houses burned down”.

Expulsions or Annihilations. In 32 cases a community was either expelled or is recorded as beingeither “destroyed” or “annihilated”. For example, the entry for Breisachamrhein reads: “Jewsare first mentioned there in 1301. The community was annihilated during theBlack Death in 1349.Subsequently, Jews again settled in Breisach but were expelled in 1424”. This indicates that thetown in question lost the economic services of the Jewish community as a result of the persecution.We also take annihilation as a measure of the intensity of persecution. Thus we distinguishbetween cases such as Passau where the Encyclopedia Judaica records “TheBlack Deathpersecutionsof 1349 caused considerable loss to the community” or Gotha where “The community suffered duringthe Black Death persecutions (1349)” from cases where the destruction of the entire community isclearly mentioned. An example of the latter is Frankfurt am Main where

“The outbreak of the Black Plague in 1349, however, changed the Jews’ protectedstatus. Jews were killed and expelled throughout Germany and Europe, and Frankfurtwas no exception. The community was completely massacred, and many Jews chose toburn down their own houses while still inside rather than face death from the angrymob.”

http://www.eea.europa.eu/data-and-maps/data/wise-large-rivers-and-large-lakes

http://www.eea.europa.eu/data-and-maps/data/wise-large-rivers-and-large-lakes


Burnings. We also distinguish cases where Jews are said to have been burned to death. Burningalive was a particularly brutal punishment reserved in medieval Europe for relapsed heretics orwomen guilty of murdering their husbands. For example, the entry for Basle (Basel) is as follows:

“During the Black Death they were accused of poisoning the wells; the membersof the city council attempted to defend them, but finally yielded to the guilds whodemonstrated before the town hall. Six hundred Jews, with the rabbi at their head,were burned at the stake; 140 children were forcibly baptized. This ended the first Jewishcommunity in Basle (Jan. 16, 1349).”

Mob Involvement. We also record whether or not the mob were listed as being the driving force inpersecuting the Jews. In Oppenhiem, for example the Encyclopedia Judaica reports that “the end ofJuly 1349, during the persecutions that followed the Black Death, most of the Jews of Oppenheimwere murdered, while others chose martyrdom (kiddush ha-Shem) and burned themselves todeath in order to escape forced conversion at the hands of the mob”. In Ulm the entry reads: “on Jan. 30, 1349, during the *Black Death persecutions, the Jewish quarter was stormed by a moband the community was all but destroyed”. In our sample there are 11 persecutions in which mobinvolvement is specifically mentioned.

Preventions. We also record whether or not there were attempts by the authorities to preventpersecutions for taking place. In Prague and Avignon, for example, the Emperor and the Popesuccessfully protected the Jews from violence. Elsewhere however, attempts at protection failed.In Cologne, the Jews had letters of protection signed by the archbishop but this did not avail them:

“Disaster overtook Cologne Jewry during the Black Death. The plague had reached thecity in the summer of 1349; the mob stormed the Jewish quarter on St. Bartholomew’sNight (Aug. 23-24), letters of protection notwithstanding. Part of the community hadassembled in the synagogue; they themselves set fire to it and perished in its flames.The rest were murdered. Among the martyrs were the last three ‘Jews’ bishops’ ofCologne (see below) and a number of distinguished rabbis.”

We find 3 persecutions that would have happened had they not been prevented by the authorities(as they reveal the demand for a persecution). Additionally, we find 8 persecutions that happeneddespite the fact that the authorities attempted to protect the community. In total, there are 11 citieswhere the authorities successfully or unsuccessfully prevented a persecution.

“Residual” Population. We estimate the residual population in the immediate aftermath of theplague in the town as the population of the town in 1300 multiplied by (100 - the Black Deathmortality rate (%) of the town)/100.

11. Sampling

Dropping selected countries. To further address concerns about sample selection, in WebAppendix Table A.9 we show that our results are robust to dropping cities in the major Europeancountries. Using modern country borders, we first drop cities in France (row 2), Germany, (row 3),


Italy (row 4), Portugal (row 5), and Spain (row 6). The size of the coefficient we obtain on mortalityremains more or less stable across specifications.

Table A.9: MORTALITY AND PERSECUTIONS, DROPPING SELECTED OBSERVATIONS




2. Drop if France Today -0.012*** [0.002] 0.984*** [0.098] 953. Drop if Germany Today -0.006** [0.003] 0.574*** [0.134] 914. Drop if Italy Today -0.007** [0.003] 0.804*** [0.114] 1045. Drop if Portugal Today -0.009*** [0.002] 0.824*** [0.106] 1186. Drop if Spain Today -0.010*** [0.002] 0.887*** [0.108] 103Notes: In rows 2-6 we drop specific countries. Robust SE’s: * p<0.10, ** p<0.05, *** p<0.01.

12. Mechanisms: Evidence on Scapegoating and ComplementaritiesIn the main text we focused on the factors that were significantly associated with eitherstrengthening or attenuating the desire to scapegoat Jews. In Web Appendix Table A.10 we reportthe results of other factors that we explored but which were not systematically related to theprobability of persecution at higher mortality rates.

Table A.10: MORTALITY AND PERSECUTIONS, OTHER INTERACTED EFFECTS



1. Close to Path of Flagellants -0.009*** [0.002] -0.010 [0.010] -0.019** [0.009]2. Close to Narbonne (Bottom 5th Pctile) -0.009*** [0.002] -0.032* [0.019] -0.041*** [0.019]3. Less Recent Entry ( 50 Years) -0.010*** [0.003] 0.003 [0.005] -0.008* [0.004]4. Cemetery, Controls for Community Size -0.012*** [0.003] 0.009 [0.006] -0.002 [0.006]5. Quarter, Controls for Community Size -0.008*** [0.003] -0.003 [0.005] -0.012*** [0.004]6. Synagogue, Controls for Community Size -0.010*** [0.003] 0.003 [0.005] -0.007** [0.005]7. Walled Density (Population ÷ Walled Area) -0.014** [0.002] 0.004 [0.008] -0.009* [0.006]8. Recent Persecution in 1321-1346 -0.010*** [0.002] 0.012 [0.008] 0.002 [0.001]9. Less Recent Persecution in 1300-1346 -0.010*** [0.002] 0.003 [0.007] -0.006 [0.006]10. Crusade Pogrom in 1147-1149 -0.009*** [0.002] 0.002 [0.010] -0.006 [0.009]11. Crusade Pogrom in 1189-1192 -0.010*** [0.002] 0.007 [0.011] -0.003 [0.011]12. Close to Alleged Ritual Murder 1st Half 14C -0.009*** [0.002] -0.002 [0.006] -0.011** [0.006]13. Close to Alleged Host Desecration 13C -0.009*** [0.002] -0.006 [0.007] -0.014** [0.007]14. Some Jews in the Town are Craftsmen or Traders -0.009*** [0.003] -0.002 [0.006] -0.011** [0.005]15. Some Jews in the Town are Doctors -0.009*** [0.002] -0.013 [0.011] -0.021** [0.010]16. Jews in the Town Contribute Special Taxes -0.009*** [0.003] 0.000 [0.005] -0.009** [0.005]

Notes: This table shows the respective effects of the mortality rate (�) and the respective interacted effects (�) of the mortality ratetimes a dummy variable shown in each row on a dummy equal to one if there has been any Jewish persecution in 1347-1352, for themain sample of 124 Jewish towns. It also shows the combination of the effect of the mortality rate (�) and the interacted effect of themortality rate and the dummy (�), to see if the interacted effect is strong effect to offset the protective effect of mortality. Robust SE’s:* p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for data sources.

Distance to Chillon Castle, Towns Warned of a Conspiracy, Rhine towns, and the Path of the


Flagellants. See the paragraph Distance to Chillon Castle, Towns Warned of a Conspiracy, the RhineRiver, and the Path of the Flagellants in Section 9..

Seat of Papacy, Bishoprics and Archbishoprics. The seat of Papacy in 1347 was Avignon. We thenobtain in GIS the Euclidean distance from each city to Avignon. We obtain data on the presence ofbishoprics and archbishoprics from a variety of sources. Bosker et al. (2013) provide informationon the locations of other bishoprics and archbishoprics for all cities in the Bosker et al. (2013)dataset. For additional bishoprics and archbishoprics we used information from Shepherd (1923).In order to ascertain that all the bishoprics in our dataset were in existence in 1300 we consultedthe following website: http://www.catholic-hierarchy.org/country/. We then create a dummyequal to one if the city was a bishopric or an archbishopric (but not for Avignon itself).

Path of the Flagellants. Historians have traditionally held the flagellants responsible formassacring Jews. However, we find no evidence that the path of the flagellant movementwas systematically associated with either persecutions or with the relationship between plagueintensity and persecution probability (see row 1 of Web Appendix Table A.10).

Distance to Narbonne. We investigate proximity to Narbonne as this is where paupers rather thanJews were blamed for the Black Death. Row 2 shows that the protective effect was accentuatedvery close to Narbonne (i.e., for towns in the bottom 5% of the Euclidean distance to it).

Recent Entrants. We use our data to create dummy variables for Jewish communities that wereestablished either 5 or 50 years prior to 1347. Note that we find no effect for cities where Jews hadentered the city in the last 50 years (row 3).

Jewish Community Characteristics. We use Berenbaum and Skolnik, eds (2007) to ascertainwhether a Jewish community had a synagogue, cemetery, or whether there was a separate Jewishquarter in a town. We ascertain whether the town was within the area of Ashkenazi settlementin the 13th century based on a map provided by (Barnavi, 1992). Rows 4-6 show that there is nodifferential effect for each characteristic considered individually when controlling for communitysize via several proxies (years of first and last entry, centrality index, dummies if persecution in1321-1346 and 1300-1346).

Walled Density. Row 7 shows that there is no differential effect across towns above and belowmedian walled density.

Recent Persecutions. We use data on past persecutions from our data set. Rows 8-9 show thiseffect is weaker if the past persecution dummy is defined for a past period of 25 years in 1321-1346(0.012, significant at 15%), and nil if it is defined for a past period of about 50 years in 1300-1346.

Crusader Pogroms, Ritual Murder, and Host Desecration. We obtain a map of both the pogromsassociated with the 1st Crusade from Beinart (1992). In the paper we report results using 1stCrusade pogroms. We also collected data on the pogroms that took place during the subsequentCrusades (1147-1149 and 1189-1192) but the later Crusades were not in general characterized bylarge-scale antisemitic violence (see rows 10-11). We find a strong positive effect for the crusade

http://www.catholic-hierarchy.org/country/


of 1189-1192, but the effect is not significant. Beinart (1992) is also our source for the locationof alleged ritual murder accusations and accusations of host desecration. It provides separatemaps for both ritual murder and host desecration allegations as well as the century in which theaccusation occurred. Rows 12-13 show the non-results for the charges of ritual murder in the firsthalf of the 14th century and the charges of host desecration in the 13th century.

Month of First Infection. As described above, we use data from Christakos et al. (2005) to obtaindata on the spread of the Black Death. We then create dummies equal to one if the month of firstinfection is December, January, February or March, April or May, or October.

We also investigate the effect for each month individually, by regressing the persecution dummyon the mortality rate, 12 “month of first infection” dummies, and the 12 interactions of thesedummies with the mortality rate. The individual and interacted effects of one month, in our caseJune, are omitted. The coefficient of the mortality rate then captures the effect of mortality in June,whereas the interacted dummies capture the relative effect of mortality in the 11 other monthscompared to June. By adding the effect for June and the interacted effects, we then obtain theabsolute effect of each month, and finally test whether this effect is significantly different fromour baseline effect of -0.009*** (row 1 of Table 1). The effects shown in Web Appendix Figure A.6confirm that the protective effect of mortality was accentuated in February, March and December,and attenuated in January, May (only significant at 15% now) and October. Note that there is noclear effect for April now, but remember that Easter took place in late April in 1348.

Figure A.6: Effects of the Black Death Mortality Rate by Month of First Infection

Notes: This figure shows for each month of first infection the effect of mortality (%) on persecution probability. Theseeffects are conditional on the individual effects of the month of first infection on the persecution dummy. We testwhether each coefficient is significantly different from the average effect across all months, i.e. the baseline effect of-0.009***: Robust SE’s: † p<0.15, * p<0.10, ** p<0.05, *** p<0.01. See Web Appendix for more details on data sources.

Financial Competition and Moneylending. We calculate distance to Cahors, Florence, Genoa,Milan, Sienna and Venice, which played important roles as major centers of moneylending in the


middle ages. We also read each entry in the Encyclopedia Judaica to see if Jewish moneylending ismentioned in the entry.

Market Access. For information on how we construct overall market access, see paragraph MarketAccess to All 1,869 Towns in Section 7..

Population, Market Access, Coast, Rivers, Medieval & Roman Roads, Hanseatic League. Forinformation on how we construct these variables, see the related paragraphs in Section 7..

Jewish Centrality Index. For information on how we construct this variable, see the relatedparagraphs in Section 8..

Holy Roman Empire. We use the shapefile in Nussli (2011) to obtain political boundaries forEurope in 1300. We then assign each city in the dataset to its political boundary in 1300 by hand.We create a dummy variable for cities that belong to the Holy Roman Empire.

Occupations. From Berenbaum and Skolnik, eds (2007), as well as Adler and Singer, eds (1906)and other sources for individual cities, we collect information on whether Jews were mentionedas craftsmen, doctors or traders in the town circa the Black Death. Rows 14-15 of Web AppendixTable A.10 show no differential effect of mortality in these towns.

Taxation. From Berenbaum and Skolnik, eds (2007), as well as Adler and Singer, eds (1906) andother sources for individual cities, we collect information on whether Jews paid special taxes inthe town circa the Black Death. Row 16 shows no differential effect of mortality in these towns.

13. Jewish Presence, Black Death Persecutions and Town Growth

Jewish Presence and Town Growth. Web Appendix Table A.11 shows that the baseline effect ofJewish presence on town population growth holds if we perform various robustness checks. Row1 reproduces the baseline results (see row 1 of Table 10). The other rows show that results hold ifwe include two lags of population (row 2), drop the towns not belonging to the original Bairochsample (row 3), and do not replace by 0.5 (500 inh.) the population of the towns with a missingpopulation (row 4), presumably because they were below the 1,000 threshold used by Bairoch (ornot in his sample because they never passed the 1,000 threshold).

Black Death Persecutions and Town Growth. In Web Appendix Table A.12, we show the year-specific effects of the Black Death persecutions (1347-1352) on town population growth, using oursample of 1,869 towns ⇥ 9 periods = 16,821 observations. The regression is the same as in Panel Bof Table 10 except we now interact the Black Death persecution dummy and presence dummy withyear fixed effects. This allows us to measure the effects of the persecution dummy for each period.Since we need to omit one period, we omit the century of the Black death, so the year 1400 for the14th century. Note that the regression includes town fixed effects and year fixed effects, the Jewishpresence dummy in each period, and the extrapolated Black Death mortality rate interacted withyear fixed effects. Column (1) shows the effects being mostly negative from 1500 (i.e. the 15thcentury). In column (2), we interact the Black Death pogrom dummy, the Black Death expulsion


Table A.11: JEWISH PRESENCE AND TOWN POPULATION GROWTH


Effect of Jewish Presence in [t-1; t] Dummy Coeff. SE Obs.

1. Baseline: Effect of Jewish Presence Dummy 0.33*** [0.04] 16,8212. Row 1 Including Two Lags of Log Town Population 0.22*** [0.03] 16,821

3. Row 1 Dropping the Non-Bairoch Towns 0.32*** [0.04] 16,128

4. Row 1 Not Replacing Missing Population by 500 Inh. 0.32*** [0.05] 9,093

Notes: The main sample consists of 1,869 towns ⇥ 9 periods = 16,821 observations. The regressions always include townfixed effects and year fixed effects. Row 1: Baseline effect (see row 1 of Table 10). Row 2: We include two lags of the townpopulation in years t-1 and t-2. Row 3: We drop the towns that do not belong to the original Bairoch sample (these townsonly belong to the Christakos sample). Row 4: We do not replace by 0.5 (500 inh.) the towns with a missing population.Robust SE’s clustered at the town level: *** p<0.01, ** p<0.05, * p<0.10. See Web Appendix for data sources.

Table A.12: BLACK DEATH PERSECUTIONS AND TOWN POPULATION GROWTH


(1) Year-Specific Effects of (2) Year-Specific Effects ofPersecution in 1347-1352 Pogrom and Expulsion in 1347-1352

Pogrom Expulsion

Effect in 1200 (i.e. 1100-1200) -0.25* [0.14] -0.24 [0.16] 0.02 [0.19]Effect in 1300 (i.e. 1200-1300) -0.07 [0.09] -0.05 [0.12] -0.12 [0.14]Effect in 1500 (i.e. 1400-1500) -0.21** [0.09] -0.30*** [0.09] 0.25** [0.12]Effect in 1600 (i.e. 1500-1600) -0.38*** [0.12] -0.45*** [0.12] 0.17 [0.15]Effect in 1700 (i.e. 1600-1700) -0.35*** [0.13] -0.44*** [0.13] 0.17 [0.18]Effect in 1750 (i.e. 1700-1750) -0.34** [0.13] -0.44*** [0.14] 0.23 [0.20]Effect in 1800 (i.e. 1750-1800) -0.37*** [0.14] -0.47*** [0.14] 0.21 [0.19]Effect in 1850 (i.e. 1800-1850) -0.17 [0.15] -0.33** [0.16] 0.31 [0.20]

Observations 16,821 16,821

Notes: Our dependent variable is log town population in year t. We interact dummies equal to one if there has been apersecution/pogrom/expulsion during the Black Death period (1347-1352) with year fixed effects (column (1): only thepersecution dummy; column (2) both the pogrom and expulsion dummies). We do not show the effect of each individualdummy. Robust SE’s clustered at the town level: *** p<0.01, ** p<0.05, * p<0.10. See Web Appendix for data sources.

dummy and the Black Death presence dummy with year fixed effects. This allows us to measurethe year-specific effects of both pogroms and expulsions. Column (2) confirms the (much) morenegative effects for pogroms over time. No negative effect is found for expulsions.

REFERENCESAbulafia, David, “The King and the Jews—the Jews in the Ruler’s Service,” in Christoph Cluse, ed., The Jews of Europe

in the Middle Ages (Tenth to Fifteenth Centuries), Vol. 4 Brepols Turnhout, Beligum 2002, pp. 43–55.Adler, Cyrus and Isidore Singer, eds, The Jewish Encyclopedia Funk and Wagnalls New York 1906.Bairoch, Paul, Cites and Economic Development, from the dawn of history to the present, Chicago: University of Chicago

Press, 1988. translated by Christopher Braider.Barnavi, Eli, The Historical Atlas of the Jewish People, New York: Schocken Books, 1992.Baron, Salo, A Social and Religious History of the Jews, Vol. IX: Under Church and Empire, New York: Columbia University

Press, 1965., A Social and Religious History of the Jews, Vol. X: On the Empire’s Periphery, New York: Columbia University Press,1965., A Social and Religious History of the Jews, Vol. XI: Citizen or Alien Conjurer, New York: Columbia University Press,1967.


Baron, Salo W., A Social and Religious History of the Jews, Vol. XII: Economic Catalyst, New York: Columbia UniversityPress, 1967.

Beinart, Haim, Atlas of Medieval Jewish History, Jerusalem: The Israel Map and Publishing Company, Ltd., 1992.Benbassa, Esther, The Jews of France, Princeton, New Jersey: Princeton University Press, 1999. translated by M.B.

DeBevoise.Benedictow, Ole J., The Black Death 1346–1353: The Complete History, Woodbridge: The Boydell Press, 2005.Berenbaum, Michael and Fred Skolnik, Encyclopedia Judaica, 2nd ed., USA: Macmillan Reference, 2007.

and , eds, Encyclopedia Judaica Macmillan Reference New York 2007.Biraben, Jean-Noel, La Peste dans l’Histoire: Les Hommes et la Peste en France et dans les Pays Europeens et Mediterraneens,

Vol. I, Mouton, Paris, 1975.Boerner, Lars and Battista Severgnini, “Epidemic Trade,” 2014. LSE Economic History Working Paper No. 212/2014.Bosker, Maarten, Eltjo Buringh, and Jan Luiten van Zanden, “From Baghdad to London: unravelling urban

development in Europe and the Arab world 800-1800,” Review of Economics and Statistics, 2013, 95 (4), 1418–1437.Botticini, Maristella, “New Evidence on Jews in Tuscany, CA. 1310–1435: The ‘Friends & Family’ Connection Again,”

Zakhor. Rivista di Storia degli Ebrei d’Italia, 1997, 1, 77–93.Cesaretti, Rudolf, Jos Lobo, Lus M. A. Bettencourt, Scott G. Ortman, and Michael E. Smith, “Population-Area

Relationship for Medieval European Cities,” PLOS ONE, 10 2016, 11 (10), 1–27.Chandler, Tertius, Four Thousand Years of Urban Growth, Lewiston, NY: St David University Press, 1987.Chazan, Robert, The Jews of Western Christendom, Cambridge: Cambridge University Press, 2006.Christakos, George, Richardo A. Olea, Marc L. Serre, Hwa-Lung Yu, and Lin-Lin Wang, Interdisciplinary Public Health

Reasoning and Epidemic Modelling: The Case of Black Death, Berlin: Springer, 2005.Cluse, Christoph, ed., The Jews of Europe in the Middle Ages, Turnhout, Belgium: Brepols Publishers, 2004.der Pfalz, Historisches Museum, ed., The Jews of Europe in the Middle Ages Historisches Museum der Pfalz Speyer 2005.Dollinger, Philippe, The German Hansa, London: MacMillan, 1970. translated and edited by D.S. Ault and S.H. Steinberg.Donaldson, David, “Forthcoming.“Railroads of the Raj: Estimating the impact of transportation infrastructure.”,”

American Economic Review, 2017.Emery, Ricard W., The Jews of Perpignan, New York: Columbia University Press, 1959.Erb, Karl-Heinz, Veronika Gaube, Fridolin Krausmann, Chritoph Plutzar, Alberte Bondeau, and Helmut Haberl, “A

comprehensive global 5 min resolution land-use data set for the year 2000 consistent with national census data,”Journal of Land Use Science, September 2007, 2 (3), 191–224.

Fischer, Gunther, Harrij van Nelthuizen, Mahendra Shah, and Freddy Nachtergaele, Global Agro-Ecological Assessmentfor Agriculture in the 21st Century: Methodology and Results, Rome: Food and Agriculture Organization of the UnitedNations, 2002.

Foa, Anna, The Jews of Europe after the Black Death, Berkeley: University of California Press, 2000.Golb, Norman, The Jews in Medieval Normandy, Cambridge: Cambridge University Press, 1998.Gottfried, Robert S., The Black Death: Natural and Human Disaster in Medieval Europe, New York: The Free Press, 1983.Jarvis, A., H.I. Reuter, A. Nelson, and E. Guevara, “Hole-filled seamless SRTM data V4,” 2008. International Centre for

Tropical Agriculture (CIAT).Jebwab, Remi, Noel D. Johnson, and Mark Koyama, “Bones, Bacteria and Break Points: The Heterogeneous Spatial

Effects of the Black Death and Long-Run Growth,” June 2016. Working Paper.Jordan, William Chester, The French Monarchy and the Jews: From Philip Augustus to the Last Capetians, Philadelphia:

University of Pennsylvania Press, 1989.Jordon, William Chester, “Home Again: The Jews in the Kingdom of France, 1315–1322,” in F.R.P. Akehurst and

Stephanie Cain Van D’Elden, eds., The Stranger in Medieval Society, University of Minnesota Press Minnesota 1997,pp. 27–46.

Kisch, Guido, The Jews in Medieval Germany: A Study of Their Legal and Social Status, Chicago: Chicago University Press,1949.

Klein, Elka, Jews, Christian Society, and Royal Power in Medieval Barcelona, Ann Arbor: The University of Michigan Press,2006.

Levenson, Alan T., The Wiley-Blackwell History of Jews and Judaism, London: Wiley-Blackwell, 2012.Luterbacher, J., D. Dietrich, E. Xoplaki, M. Grosjean, and H. Wanner, “European seasonal and annual temperature

variability, trends, and extremes since 1500,” Science, 2004, 303, 1499–1503.McCormick, Michael, Origins of the European Economy, Cambridge: Cambridge University Press, 2001.McEvedy, Colin and Richard Jones, Atlas of World Population History, New York: Facts of File, 1978.Nahon, GErard, “Zarfart: Medieval Jewry in Northern France,” in Christoph Cluse, ed., The Jews of Europe in the Middle

Ages (Tenth to Fifteenth Centuries), Vol. 4 Brepols Turnhout, Beligum 2002, pp. 205–220.Nussli, Christos, 2011. URL: http://www.euratlas.com/about.html.Pounds, Norman J. G., An Historical Geography of Europe: 450 B.C.–A.D. 1330, Cambridge: Cambridge University Press,

1973.Pryor, J. H., Geography, Technology, and War: Studies in the Maritime History of the Mediterranean, 649-1571., Cambridge:

Cambridge University Press, 1992.Quesada, Miguel Angel Ladero, “Castile: an overview (Thirteenth to Fifteenth Centuries),” in Christoph Cluse, ed., The

Jews of Europe in the Middle Ages, Brepols Publishers 2004, pp. 151–161.


Ramankutty, N.J., J. Foley, J. Norman, and K. McSweeney, “The Global Distribution of Cultivable Lands: CurrentPatters and Sensitivity to Possible Climate Change,” Global Ecology and Biogeography, 2002, 11, 377–392.

Ries, Rotraud, “German Territorial Princes and the Jews,” in R. Po-Chia Hsia and Hartmut Lehmann, eds., In and Out ofthe Ghetto: Jewish-Gentile Relations in Late Medieval and Early Modern Germany, Cambridge University Press Cambridge1995, pp. 215–246.

Roosen, J. and D.R. Curtis, “Dangers of Noncritical Use of Historical Plague Data,” Emerging Infectious Diseases, 2018,Forthcoming.

Roth, Cecil, “Genoese Jews in the Thirteenth Century,” Speculum, 1950, 25 (2), 190–197., “The Economic History of the Jews,” The Economic History Review, 1961, 14, 131–135.

Roth, Front Cover Norman, Medieval Jewish Civilization: An Encyclopedia, London: Routledge, 2014.Russell, Josiah Cox, Medieval Regions and their Cities, Newton Abbot, Devon: David & Charles, 1972.Schmid, Boris V., Ulf Buntgen, W. Ryan Easterday, Christian Ginzler, Lars Walløe, Barbara Bramanti, and Nils Chr.

Stenseth, “Climate-driven introduction of the Black Death and successive plague reintroductions into Europe,”Proceedings of the National Academy of Sciences, 2015, 112 (10), 3020–3025.

Schwarzfuchs, Simon R., “The Expulsion of the Jews from France (1306),” The Jewish Quarterly Review, 1967, 57, 482–489.Segre, Renata, The Jews in Piedmont, Vol. 1, Jerusalem: The Israel Academy of Sciences and Humanities and Tel Aviv

University, 1986.Shank, Michael H., Unless You Believe, You Shall Not Understand: Logic, University, and Society in Late Medieval Vienna,

Princeton, New Jersey: Princeton University Press, 1988.Shatzmiller, Joseph, “Les Juifs de Provence pendant la Peste Noire,” Journal of Jewish Studies, 457–480 1974, 133.Shepherd, William R., Historical Atlas, New York: Henry Holt and Company, 1923.Spector, Shmuel and Geoffrey Wigoder, eds, The Encyclopedia of Jewish Life Before and During the Holocaust, New York:

NYU Press, 2001.Stasavage, David, “Was Weber Right? The Role of Urban Autonomy in Europe’s Rise,” American Political Science Review,

5 2014, 108, 337–354.Taitz, Emily, The Jews of Medieval France: The Community of Champagne, Westport, Conn.: Greenwood Press, 1994.Talbert, Richard J. A, ed., Barrington Atlas of the Greek and Roman World Princeton University Press Princeton 2000.Toch, Michael, ed., Peasants and Jews in Medieval Germany Ashgate Aldershot 2003.Voigtlander, Nico and Hans-Joachim Voth, “The Three Horsemen of Riches: Plague, War, and Urbanization in early

modern Europe,” Review of Economic Studies, 2013, 80, 774–811.Zanden, Jan Luiten Van, Eltjo Buringh, and Maarten Bosker, “The Rise and Decline of European Parliaments, 1188–

1789,” Economic History Review, 08 2012, 65 (3), 835–861.Ziegler, Philip, The Black Death, London: Collins, 1969.

negative shocks and mass persecutions: evidence from the...

Documents